Skip to contentAI articles
AI NewsClaude Fable 5.2: 20% price cut, up to 30% fasterAnthropic released Claude Fable 5.2 on November 9, 2026. API prices fall 20% to $8/$40 per million tokens, and output is up to 30% faster than Fable 5.1.#Claude#Anthropic#LLM Updates
GuidesAI fact check: only 40% of cited stats hold upAI fact check: Trakkr traced 1,285 stats cited by ChatGPT, Perplexity, Gemini, Claude and AI Overviews. Only 40% led to a matching study. How to check one.#Fact checking#ChatGPT#Perplexity
AI NewsApple Intelligence launches in China with QwenApple Intelligence comes to iPhones in China with iOS 27.3 in December. Alibaba's Qwen handles some Siri and Writing Tools requests, but Siri AI is left out.#Apple Intelligence#Qwen#Siri
AI NewsFlash prices up 5x on AI, eroding Apple's leverageAI demand made flash 5x pricier in a year, the WSJ reports. Cloud giants now buy twice as much as Apple, and doubling iPhone 17 Pro storage costs Apple $66.#Semiconductors#Apple#Memory
AI NewsMeta's Style Sketch builds outfits from InstagramMeta's free Style Sketch app builds a closet from your Instagram photos and suggests outfits. It is out only in Indonesia, Singapore and Thailand, not the US.#Meta#Instagram#Fashion
AI NewsOpenAI, ServiceNow expand deal with ChatGPT pluginOpenAI and ServiceNow expanded their deal on Oct 28, 2026. A ServiceNow MCP plugin for ChatGPT is live now; Codex, security and voice tools follow in 2027.#ChatGPT#OpenAI#ServiceNow
AI NewsProton cancels AI re-interviews, keeps pushing AIProton called off its AI re-interviews of staff on Oct 30 after 404 Media reports. CEO Andy Yen admitted a communication failure but will keep pushing AI.#Proton#AI adoption#AI coding
GuidesVoiceStudio: free local ElevenLabs alternativeVoiceStudio is a free, open-source app that clones voices and transcribes speech on your own computer. Its default OmniVoice model is non-commercial.#VoiceStudio#Text to speech#Voice cloning
AI NewsWaymo will test Toyota bZ4X robotaxis in JapanToyota will build bZ4X robotaxis for Waymo at its Takaoka plant, with testing in Japan from 2027 and the UK to follow. No city, fleet size or fare is set yet.#autonomous driving#Waymo#Toyota
AI NewsReddit to require logins as search referrals fallReddit will test a logged-in-only site outside the US in Q4. Search referrals fell over 1M a week in October; its CEO says Google's AI Mode drives no traffic.#Reddit#Google#AI Mode
GuidesAI crawler training opt-out in one robots.txt ruleAI crawler training opt-out now takes one robots.txt rule. OpenAI, Google and Microsoft will honor it, Microsoft from Feb. 1, 2027. Until then, name each bot.#robots.txt#AI crawlers#OpenAI
AI NewsAmazon's Jassy: shopping agents 'still not great'On Amazon's Oct 29 earnings call, Andy Jassy said outside AI shopping agents aren't hurting its ads. Amazon blocked Meta's Muse and backs Alexa for Shopping.#Amazon#Meta#AI agents
AI NewsChatGPT and Codex usage limits merge in DecemberFrom December, Plus, Pro and Business plans will share one token-based allowance across ChatGPT and Codex. Instant chat won't count, and no start date is set.#ChatGPT#Codex#OpenAI
AI NewsChatGPT now works with iCloud Mail and CalendarOpenAI added iCloud Mail and Calendar plugins to ChatGPT on Oct 27 for all plans. You connect with an Apple app-specific password that outlives a disconnect.#ChatGPT#OpenAI#iPhone
AI NewsClaude Code rule blocks recreating existing gamesA server-side rule stops Claude Code on Opus 5.5 from recreating existing games. Anthropic confirmed it on X, and a wargame studio says paid work stopped.#Claude Code#Anthropic#game development
GuidesGemini create files: edit PDFs and docs in chatGemini create files: docs, PDFs, slides and sheets now open in the chat as previews you can edit. Phones edit only PDFs; the web handles the rest.#Gemini#Google#File Creation
AI NewsCopilot Fusion now picks the model for each taskGitHub made Copilot Fusion generally available Oct 28. Pick Standard, High or Max and Copilot picks the models. High and Max need Pro+. 30% off until Nov 30.#GitHub#Copilot#AI Agents
AI NewsMalaysia seizes 30 AI chip servers at KLIA airportMalaysian customs seized 30 AI chip servers worth RM49.5 million at KLIA, declared as electronic components, in the year's fifth case tied to one syndicate.#Malaysia#Export Controls#AI Chips
AI NewsNvidia's Jensen Huang close to alleged Chinese spyThe NYT says US officials judged Megaspeed's Alice Huang, close to Nvidia CEO Jensen Huang, a Chinese intelligence asset. Nvidia calls it normal business.#Nvidia#Export Controls#AI Chips
AI NewsChatGPT limits weapons information for teensOpenAI said on Oct 30, 2026 that ChatGPT will limit weapons information for teens, end chats fixated on violence, and report on safety every six months.#ChatGPT#OpenAI#AI Safety
GuidesSurvival LLM: offline AI assistant for emergenciesSurvival LLM is a free offline AI assistant that keeps a local model and 1.5 million reference articles on your computer and cites sources with no internet.#Survival LLM#Offline AI#Emergency preparedness
AI NewsUK's AISI gets £120m, about 3-4x US budget a yearThe UK Budget gives the AI Security Institute an extra £120 million over four years. Semafor says 10x the US budget; per year it is about 3 to 4 times.#AISI#AI safety#UK
GuidesWikipedia AI policy: how editors remove AI textWikipedia AI policy: since March 2026, English Wikipedia bans AI-written article text. Editors spot it by phrasing and weak sources and delete clear cases.#Wikipedia#AI-Generated Content#AI Detection
AI NewsStanford professors quit Anthropic book settlementTwo Stanford professors reportedly left Anthropic's $1.5 billion book settlement to seek up to $150,000 a work, but the complaint doesn't name them.#Anthropic#Claude#Copyright
AI NewsMicrosoft offers buyouts to about 60,000 employeesMicrosoft offers voluntary buyouts to staff at level 70 and below in four countries. CNBC says about 60,000 are eligible; those who accept leave Dec. 31.#Microsoft#Layoffs#AI investment
AI NewsAmazon's 5 new Echo devices: prices and Alexa+Amazon unveiled five Alexa devices. Four are on sale now and Echo Show 16 ships later this year. Here are the prices and which features need Alexa+.#Amazon#Alexa+#Smart speakers
GuidesAutomation bias in AI: experts followed the AI 86%Automation bias in AI is over-reliance on automated systems. In Anthropic's trial, analysts matched a coin-flip AI lean on 86% of calls yet felt in charge.#Anthropic#Automation bias#AI agents
GuidesWhat is Genie 4? Google's world model with audioGenie 4 is Google DeepMind's world model that adds ambient sound and live dialogue to explorable worlds. Only testers have it; Ultra subscribers are next.#Google#Google DeepMind#Genie 4
GuidesWhy OpenAI and Anthropic slashed LLM API pricingLLM API pricing at OpenAI and Anthropic fell by up to 80% in mid-2026. Max Woolf credits easy switching, cheap open-weight rivals and cheaper serving.#OpenAI#Anthropic#API pricing
GuidesSakana AI API Platform: Fugu with admin controlsSakana AI API Platform entered early access on October 28, 2026, offering Fugu and Fugu Ultra via an OpenAI-compatible API with audit logs and budget caps.#Sakana AI#LLM API#Enterprise AI
GuidesSiri AI supported devices and the waitlist stepsSiri AI supported devices start at iPhone 15 Pro. Here is the storage each needs, the six languages in iOS 27.2, and how to join the waitlist in Settings.#Apple#Siri#Siri AI
AI NewsAnthropic Buys One-Person Biotech Mithril MedicineAnthropic acquired Mithril Medicine, a one-person biotech that used Claude to run 11 labs. Its muscular dystrophy drug could enter human trials in late 2028.#Anthropic#Claude#Drug discovery
AI NewsAnthropic's S-1: $16.3B revenue and an $80.7B lossAnthropic's S-1 shows $16.3 billion in first-half revenue. $72.4 billion of its $80.7 billion loss is non-cash, and the founders keep 50.1% of the vote.#Anthropic#IPO#AI industry
AI NewsChatGPT ads add 26 countries, reaching 89 in totalOpenAI added 26 countries to ChatGPT Ads on October 21, for 89 in total, as advertisers topped 100,000. Ads show only on the Free and Go plans.#ChatGPT#OpenAI#AI advertising
AI NewsParents sue OpenAI over ChatGPT seizure-med adviceBowen Wan's parents sued OpenAI on October 27, saying GPT-4o tapered her seizure drug to zero and told her to hide it. OpenAI cites an older model.#OpenAI#ChatGPT#lawsuit
GuidesChrome enhanced autofill and Gmail: turn it offChrome enhanced autofill now suggests order numbers, tracking info, and flight bookings from Gmail. Who gets it, and how to turn it off or remove suggestions.#Google Chrome#Gmail#Gemini
GuidesClaude Duet: free on Pro and Max, US-only for nowClaude Duet is Anthropic's proactive agent app, a US-only research preview since October 21, 2026. It's free on Pro and Max but draws on your usage limits.#Claude#Anthropic#AI agents
AI NewsElvis rights holders sue Anthropic over 274 songsPresley's rights holders say Anthropic trained Claude on lyrics to 274 songs without a license, and seek about $48.6 million and the destruction of copies.#Anthropic#Claude#Copyright
AI NewsOpenAI's GPT-6 arrives on Google Cloud in previewGPT-6 Astra, Sol and Luna are in Google Cloud's Model Garden for select enterprises, billed via Google Cloud. Prompts aren't used for training. No price yet.#Google Cloud#OpenAI#GPT-6
AI NewsGmail, Photos linked to AI Mode can now shape adsGoogle's October 16 privacy policy lets Gmail and Photos data shape ads for people who connect them to AI Mode. Google says inboxes won't drive ads directly.#AI security#Privacy#Advertising
GuidesApple MCP in Keynote, Pages and Numbers explainedApple MCP support comes to Keynote, Pages and Numbers in November, letting outside AI agents read and edit documents on macOS 27, plus a beta assistant.#Apple#MCP#Keynote
AI NewsKimi K3.1 Pro benchmarks, pricing and open weightsKimi K3.1 Pro is Moonshot AI's post-trained Kimi K3. It nearly matches Claude Opus 5.5 on coding, costs a third less than K3 and now has open weights.#Kimi K3.1 Pro#Moonshot AI#Open-weight models
AI NewsOpenAI says AI refuted Exponential Time HypothesisOpenAI says an internal AI refuted the Exponential Time Hypothesis. Two of ETH's original proposers lead the review, and Lean verified only a weaker bound.#OpenAI#AI and math#Lean
AI NewsTesla Kills Pedestrian on I-80; Driver Cites FSDA Tesla hit and killed a 71-year-old man crossing I-80 in Vallejo on Oct. 22. The driver says FSD was on; dashcam video shows the car braking and swerving.#Tesla#FSD#Traffic Safety
GuidesWhat Is GPT-6.1 Terra? Pricing and BenchmarksGPT-6.1 Terra, OpenAI's mid-tier GPT-6 model out Oct. 26, 2026, costs $1 input and $5 output per 1M tokens, half of Sol's, and matches Astra on computer use.#OpenAI#GPT-6#ChatGPT
GuidesSakana Aquarium: An Enterprise AI Agent PlatformSakana Aquarium is Sakana AI's enterprise AI agent platform, announced Oct. 23, 2026. It maps company data in an ontology and runs agents with human approval.#Sakana AI#AI agents#Ontology
GuidesSakana Chat's My Fish shows what the AI remembersMy Fish shows what Sakana Chat's memory holds about you. Memory is on by default, opting out of training is separate, and Sakana Chat only works in Japan.#Sakana AI#Sakana Chat#AI chatbot
AI NewsDigital Hollywood University to roll out ClaudeTokyo's Digital Hollywood University will give every student and staff member Claude from November 2026, which it says is a first among Japan's universities.#Anthropic#Claude#Digital Hollywood University
GuidesGoogle Deep Review: what the AI paper checker doesGoogle Deep Review checks papers for logic gaps, math errors and weak evidence. Two reviews a week are free, and its predecessor caught 89.7% of known errors.#Google#Gemini#Research papers
AI NewsMistral Large 4 (Le Chonk) specs and benchmarksMistral Large 4, a 1.05T-parameter MoE with 52B active, went open-weight on October 29 under a modified MIT license. Input now costs $0.70 per million tokens.#Mistral Large 4#Mistral AI#Open-weight model
AI NewsAmazon's Bee Frontier and Eclipse AI note-takersBee Frontier is a $74.99 wearable on sale now; Bee Eclipse clips to your phone from $99.99 soon. What each does, what Bee Pro adds and how audio is handled.#Amazon#Alexa+#Wearables
AI NewsAnthropic Japan at one: customers up about tenfoldAnthropic Japan's staff grew from 3 to over 50 and customers about tenfold in a year. It is keen on a Japan data center but did not comment on Osaka.#Anthropic#Claude#Japan
AI NewsClaude Platform adds OpenTelemetry agent tracingClaude Platform now exports OpenTelemetry traces in public beta. Managed Agents, the Messages API and SDKs send them to Datadog, Langfuse or any OTLP tool.#Claude#AI agents#OpenTelemetry
AI NewsFTC seeks safety records from OpenAI and AnthropicOn October 14 the FTC sent OpenAI and Anthropic civil investigative demands for safety and privacy records. It is collecting evidence, not ruling.#OpenAI#Anthropic#Regulation
AI NewsGoogle is feeding rank trackers poisoned SERPsNozzle says Google has served rank trackers normal-looking SERPs with the wrong sites since mid-September. What it measured, and how to spot a fake drop.#Google#SEO#Search Console
AI NewsGrok Voice 3.0 tops speech AI index, still $0.08SpaceXSI released Grok Voice 3.0 on October 14. It leads the Artificial Analysis index at 86.9 for $0.08 a minute, but GPT-Live-1 still wins on support calls.#Grok#SpaceXSI#Voice AI
AI NewsLFM2.5-24B-A2B: Liquid AI's MoE for local agentsLFM2.5-24B-A2B is Liquid AI's 24B MoE with 2B active. It leads or matches Qwen3.6-35B-A3B on 5 of 8 agent tests and runs at 184 tok/s in 15 GB on an M5 Max.#Local LLM#AI agents#MoE
AI NewsMizuho turns non-engineers into AI developersMizuho's bootcamp trains non-engineers as AI developers in three months with Claude Code and Codex. Its target is 100 people; thousands is a Nikkei figure.#Claude Code#Codex#Reskilling
AI NewsNVIDIA's $250M compute gift for UK AI agent safetyNVIDIA will give UK researchers $250M of compute over five years for AI agent safety work, not cash. Jensen Huang also previewed $2B in new European deals.#NVIDIA#AI agents#AI safety
AI News8 states sue OpenAI, alleging ChatGPT harms kidsEight states sued OpenAI on October 14, saying ChatGPT harmed kids and misled parents. Iowa seeks up to $40,000 per violation; OpenAI cites age protections.#OpenAI#ChatGPT#lawsuit
AI NewsOpenAI to retire GPT-4o and GPT-4.1 API on Jan. 14OpenAI removes five GPT-4o and GPT-4.1 snapshots from its API on January 14, 2027, 92 days after notice. What stops, what Luna costs and what code to change.#OpenAI#GPT-4o#GPT-4.1
AI NewsAnthropic moves its monorepo from git to ScionAs Claude wrote more of its code, Anthropic's daily merges rose from about 2,500 to over 30,000, so it moved its monorepo from git to its own system, Scion.#Anthropic#Claude#git
AI NewsAnthropic opens Singapore office, signs AI MOUAnthropic opened its fifth Asia-Pacific office in Singapore on October 14, 2026, and signed an MOU with the government on AI safety, adoption and skills.#Anthropic#Claude#AI safety
AI NewsApple HomeView brings Siri AI to a $299 home hubApple HomeView is a $299 home hub with a 7-inch screen that recognizes up to six faces and runs Siri AI. It ships November 13, but not in the EU.#Apple#Siri#Smart home
GuidesEmbedding Gemma Transcribe: speed and accuracyEmbedding Gemma Transcribe turns Google's EmbeddingGemma 2 into a free Mac speech to text app: an hour of English in 12.6 s, at about 5% word error rate.#EmbeddingGemma#speech to text#dictation
GuidesREA: AI Reverse Engineering MCP for Claude CodeREA is a free AI reverse engineering MCP server for Claude Code and Codex. One npx command sets it up, and Ghidra, IDA or Hopper reads apps without source.#REA#MCP#Reverse engineering
AI NewsCharacter.AI sale: 39 states want chat safeguardsCharacter.AI filed for Chapter 11 on October 12, 2026. Two days later, 39 states and DC asked any buyer of its user data to adopt nine protections.#Character.AI#Chapter 11#Disney
AI NewsEpic Games delays Disney shooter, goes AI firstBloomberg reports Epic Games told its Marvel and Star Wars shooter team to stop hand-writing code and use AI. The March 2027 release is off with no new date.#Generative AI#Game development#AI coding
AI NewsMeta Muse Wonder: benchmarks, pricing, Wonder FullMeta's Muse Wonder API costs $4/$20 per million tokens, the same as Claude Opus 5.5. Wonder Full ties Opus on Meta's table but is only in Muse Code.#Meta#Muse Wonder#AI agents
AI NewsMicrosoft to deploy Maia 300 AI chips from 2027Microsoft will put its Maia 300 AI chip into production in 2027. It expects 40% more tokens per megawatt than leading GPU systems but has not named them.#Microsoft#Azure#AI chips
GuidesFree live transcription on your PC and phoneLive transcription turns speech into text as people talk. Do it for free with Google Docs, Windows 11 Live Captions, Live Transcribe, or iPhone Voice Memos.#transcription#speech recognition#voice typing
AI NewsThe X Subscription: Grok, Cursor and X from $8SpaceXSI put Grok, Grok Bot, Cursor and X Premium on one bill on October 12, from $0 to $200 a month. Claude, GPT and Grok API use cost extra.#Grok#Cursor#X
AI NewsarXiv caps new paper submissions at two per monthSince October 1, 2026, arXiv limits each submitter to two new papers a calendar month. Rejections count; co-authors don't. September hit a record 40,363.#arXiv#research papers#AI policy
GuidesFree OBS captions on CPU only: setting up HayamimiFree OBS captions on CPU alone: Hayamimi is an MIT-licensed speech recognizer that runs in under 2GB of RAM. Set it up in OBS and learn its limits.#speech recognition#captions#OBS
GuidesWhat is knowledge distillation? Smaller AI modelsKnowledge distillation trains a small student model to copy a large teacher. Here's how it works, what DeepSeek did, and when it breaks AI terms of service.#knowledge distillation#AI models#LLM
GuidesWhat is mixture of experts (MoE)? How it worksMixture of experts (MoE) splits an AI model into many experts and runs only a few per token. See how routing works, what active parameters mean and the catch.#mixture of experts#MoE#LLM
GuidesWhat is an open weight model? AI you can downloadAn open weight model is an AI model whose trained weights anyone can download and run. It isn't open source, and its license decides what you can do with it.#open weight model#open source AI#LLM
GuidesWhat is LLM quantization? 4-bit vs 8-bit explainedLLM quantization stores a model's weights in fewer bits, such as 4 instead of 16, to cut memory. Learn how it works, what Q4_K_M means and how accuracy drops.#quantization#local LLM#LLM
GuidesHow to turn off Apple Intelligence on macOS 27On macOS 27 you can turn off Apple Intelligence features but not delete the models. How to remove them with RemoveMacAI, what breaks, and how to undo it.#Apple Intelligence#macOS#open source
GuidesAI Image Detection: 3 Free Tools to Spot AI PhotosStart AI image detection with free tools: read the file's records, check for a SynthID watermark and get a Hive score. Trace the source only if they disagree.#AI images#AI detection#AI-generated content
GuidesWhat is a context window? How much an AI can readA context window is the text, in tokens, an AI reads for one reply. Prompts, chat history, files and the reply all count. When it fills, old messages drop.#context window#tokens#LLM
AI NewsClaude Code hooks: 5 types and the node spawn costClaude Code hooks come in five types, from command to agent. As command hooks, one Edit launched 17 node processes, and hooks drove 47–87% of node launches.#Claude Code#Anthropic#Hooks
AI NewsMicrosoft-Decision-1: Jev-priced decision modelMicrosoft-Decision-1 scores fixed options instead of writing text, at Jev's price of $0.042 per million input tokens. JevBench measured 459 ms, not 85 ms.#Microsoft#Microsoft Foundry#Decision models
AI NewsPC shipments fell 20.1% in Q3 as prices stay highIDC says worldwide PC shipments fell 20.1% to 62.7 million in Q3 2026 after a first-half stock-up. AI data center demand keeps prices high, and HP fell 30.9%.#PC#IDC#Memory
AI NewsAnthropic usage policy update: what changes Nov 12The Anthropic usage policy update takes effect Nov. 12. Venting at Claude is fine, while fake accounts, armed drones and covert tracking are now spelled out.#Anthropic#Claude#Usage policy
AI NewsClaude Dynamic Workflows come to Managed AgentsClaude Dynamic Workflows are now in public beta in Claude Managed Agents. One run can start up to 1,000 agents, 64 at a time, billed at each model's rates.#Claude#Claude Code#AI agents
GuidesWhat is the Gemini agent? Features and pricingThe Gemini agent is Google Cloud's AI agent for work. It runs inside Gmail, Docs and Slack, builds coworker agents and picks Gemini or Claude models per task.#Gemini#Google Cloud#Google Workspace
AI NewsOpenAI Trims Round to $20B at a $1.25T ValuationOpenAI now seeks up to $20B, down from $30B, at a $1.25T post-money valuation per the FT. September's $1.4T was pre-money, so the two don't compare directly.#OpenAI#IPO#AI industry
AI NewsOpenAI math papers withdrawn over one sign errorOpenAI math papers: 3 of 722 AI-written papers were pulled a day after release. One sign error broke a proof two others relied on. About 42% are Lean-checked.#OpenAI#AI and math#Lean
AI NewsSynthID Detector: How to Check for AI WatermarksGoogle opened SynthID Detector to everyone on October 7, 2026. How to check a file for its AI watermark, and why a clean result doesn't prove a human made it.#Google#SynthID#AI detection
AI NewsTrump Super Intelligence Rename: Who Must Use SITrump calls anyone who says artificial intelligence THE ENEMY, but his super intelligence executive order covers only federal agencies' documents, not firms.#Donald Trump#Super Intelligence (SI)#AI policy
AI NewsJev maker TypeSafe AI raises $870M at $7.5BTypeSafe AI raised $870M at a $7.5B valuation 24 days after launching Jev. Five days later Jev 1.14 added images, box and long context as experiments.#TypeSafe AI#Jev#Decision models
AI NewsCloudflare Clef vs Jev: decision model benchmarksCloudflare Clef is an open-weight decision model that speaks Jev's API and reads images, at about 6x Jev's price. Its benchmark lead is still self-reported.#Cloudflare#Workers AI#Decision models
AI NewsToshiba hard drive capacity set to nearly doubleToshiba hard drive capacity is set to nearly double by FY2027, measured in storage, not units. WDC and Seagate stock closed about 10% lower on October 2.#Toshiba#HDD#Data center
AI NewsClaude Haiku 5.5 API pricing and benchmarksClaude Haiku 5.5 costs a tenth of Haiku 4.5's per-token price on prompts up to 100K. See the API price table, the 100K catch, and benchmarks vs Sonnet 5.5.#Claude#Anthropic#LLM updates
AI NewsGoogle Docs Markdown: edit .md files nativelyGoogle Docs Markdown support now opens and edits .md files without converting them, and Drive previews them. Rollout began October 5, 2026, for all accounts.#Google Docs#Google Workspace#Markdown
AI NewsClaude Google Sheets, Docs and Slides add-on guideThe Claude Google Sheets add-on, Claude for Google Workspace, edits Sheets, Docs and Slides files in place on paid plans. Setup, edit modes and beta limits.#Claude#Anthropic#Google Workspace
AI NewsGPT-6 in ChatGPT: release dates and Intelligent UIGPT-6 reached ChatGPT on October 7, 2026: Sol for paid plans, Luna for free and Go from October 8, and Intelligent UI answers with charts and buttons.#OpenAI#GPT-6#ChatGPT
AI NewsEmbeddingGemma 2: open multimodal embedding modelEmbeddingGemma 2 maps text, images, audio and video into one vector space, runs on-device in ~191MB of RAM, and works with Ollama for local RAG.#EmbeddingGemma#Gemma 4#Embedding models
AI NewsChatGPT ads come to image generation in a US testChatGPT ads get a visual format: OpenAI tests it in image generation for US Free and Go users from late October 2026. Plus and Pro stay ad-free.#ChatGPT#OpenAI#AI advertising
AI NewsChatGPT watermark: EU only, and edits weaken itThe ChatGPT watermark, textGrain, now goes into ChatGPT text in the EU only. Who gets it, why AI detectors cannot read it, and how much editing weakens it.#ChatGPT#OpenAI#Watermarking
AI NewsGemini free tier goes Flash-Lite only on October 9The Gemini free tier is Flash-Lite only from October 9, 2026. Flash needs AI Plus ($4.99/month), Pro needs AI Pro ($19.99), and Deep Think stays Ultra-only.#Gemini#Google
AI NewsReflection Beam is a 501B open-weight MoE modelReflection Beam has 501B total and 23B active parameters. See its scores against Kimi K3, GLM 5.2 and DeepSeek V4 and when the Apache 2.0 weights ship.#Reflection Beam#Reflection AI#Open-weight models
AI NewsMeta Muse Keeps Hourly Profiles of People You KnowExtracted system prompts show Meta Muse keeps an hourly page on your family, friends and coworkers. What the pages hold, Meta's reply, and how to delete them.#Meta#AI Agents#Privacy
AI NewsClaude Sonnet 5.5 vs Opus 5.5: price, benchmarksClaude Sonnet 5.5 costs half as much per token as Opus 5.5 and beats it on Terminal-Bench 4.0. Where Opus leads, when half price holds, and 5 API fixes.#Claude#Anthropic#LLM updates
AI NewsAMD ROCm 10: ROCm.AI Brings Skills, Hyperloom, CLIROCm 10 makes AMD's ROCm.AI generally available. What AMD Skills, Hyperloom, and the ROCm CLI do, and how to read the 3.3x claim for Radeon, Windows, CUDA.#AMD#ROCm#GPU
AI NewsAnthropic Pentagon rulings split; Claude droppedA judge voided the Anthropic Pentagon label Aug. 27, the D.C. Circuit upheld a second one Sept. 25, and the Pentagon told the BBC it stopped using Claude.#Anthropic#Claude#AI Policy
AI NewsClaude Opus 5.5 pricing, benchmarks, API changesClaude Opus 5.5 API costs $4 input and $20 output per million tokens, 20% below Opus 5, and matches Fable 5.1 on most work. Four API changes break old code.#Claude#Anthropic#LLM Updates
GuidesGemini Notebook (formerly NotebookLM) explainedGemini Notebook is what NotebookLM became on July 16, 2026. Free users get 50 chats and 3 Audio Overviews a day and it now reads ebooks bought on Google Play.#Gemini Notebook#NotebookLM#Gemini
AI NewsClaude in Chrome vs the Cowork built-in browserClaude in Chrome or the Cowork built-in browser? Both shipped on August 26, 2026. Which one fits which task, what the safety numbers say, and how to switch.#Claude#Claude Cowork#Claude in Chrome
AI NewsNvidia Hugging Face Deal Draws DOJ QuestionsThe DOJ is asking whether Nvidia's $12.93 billion Hugging Face deal would hurt AI competition, Reuters reported. Closing is due in the first half of 2027.#NVIDIA#Hugging Face#Open Source
AI NewsGemini 3.5 Transcribe: pricing and API limitsGemini 3.5 Transcribe is Google's speech to text API at about $0.005 per minute recorded and $0.009 live. What the two models charge and where they stop.#Gemini#Google#Speech recognition
AI NewsGPU Utilization: Adobe's GPUs Sat Two-Thirds IdleLow GPU utilization at Adobe traced to slow storage and uneven work per GPU. Per Backblaze, two-thirds of GPU time went to waiting; fixes reached about 80%.#GPU#AI Training#Infrastructure
GuidesAGENTS.md Best Practices: Keep Rules From FadingAGENTS.md is an open format for guiding coding agents. See an example, how it differs from CLAUDE.md, and why its rules fade in long sessions.#AGENTS.md#AI Agents#Coding Standards
AI NewsAnthropic Nscale: $45B Compute Deal ReportedAnthropic Nscale reportedly signed a $45B compute rental on Nvidia Vera Rubin chips from late 2027. See the terms, the source and what is unconfirmed.#Anthropic#Nscale#AI Infrastructure
AI NewsAnthropic Grants: $5M for AI Wellbeing ResearchAnthropic grants total $5M for independent research on how AI affects wellbeing. Grantees build open-source evaluations. Application deadline: September 21.#Anthropic#AI Safety#Wellbeing
AI NewsMac Studio M5 Ultra: 512GB, Price and Release DateMac Studio M5 Ultra price starts at $5,499, with arrivals from September 22 and the 512GB build in late October. See specs, bandwidth and Apple's benchmarks.#Apple#Mac Studio#Local LLM
AI NewsWebMCP: What It Is and How ChatGPT Supports ItWebMCP is an experimental standard that lets sites publish tools AI agents call. ChatGPT's in-app browser supports it; OpenAI offered the top 10 $3,000 each.#WebMCP#OpenAI#AI Agents
AI NewsClaude Fable 5 Back as FT Says Top Model StrugglesThe FT says Anthropic's best model struggles to attract users. Claude Fable 5 was suspended June 12, returned July 1 and moved to usage credits July 7.#Claude Fable 5#Anthropic#AI Models
AI NewsKiro AI Agent Incident Triage: A 13m 35s DiagnosisAI agent incident triage at Kiro: a 2:33 AM alarm was diagnosed in 13 minutes 35 seconds. A representative month saw about 250 unattended investigations.#Kiro#AI Agents#Operations
AI NewsMetaRoCE: Meta's RDMA over Ethernet for AI, No PFCMetaRoCE is Meta's RDMA transport for AI on Ethernet with no PFC. It held about 86% throughput at 1% packet loss, and the spec goes to OCP in October 2026.#MetaRoCE#Meta#AI Infrastructure
AI NewsNVIDIA AVO Scores 100 on the ARC-AGI-3 Public SetNVIDIA AVO scored 100.00 RHAE on all 25 ARC-AGI-3 public-set environments, NVIDIA says. It credits the harness, and the jump from 30 isn’t a controlled test.#NVIDIA#AI Agents#ARC-AGI-3
GuidesImage Generation Model Comparison on One PromptAn image generation model comparison from OpenRouter: 44 listings, 42 scored out of 4, with prices of $0.006 to $0.25 and times of 1.2s to 208.2s.#OpenRouter#Image Generation#Benchmarks
AI NewsWindows Paint Invisible Watermark: A Server GUIDA researcher found Paint and Photos embed an invisible 16-byte GUID in AI images. It comes from Microsoft's server and reappears in the C2PA manifest.#Microsoft#Watermarking#C2PA
AI NewsPerplexity Portable Computer Runs a Local AI AgentPerplexity Portable Computer, launched August 25, 2026, runs an AI agent on an NVIDIA DGX Spark. On-device work uses no credits; cloud handoffs ask first.#Perplexity#NVIDIA#On-Device AI
AI NewsLiquid AI Pipette: Mobile AI Benchmark for PhonesLiquid AI released Pipette, an open-source mobile AI benchmark that measures model, quantization, runtime, and device together on real phones and laptops.#Pipette#Liquid AI#Benchmarks
AI NewsQwen 3.8 Flash Next: 125B Qwen4 Preview, 6B ActiveQwen 3.8 Flash Next activates 6B of 125B parameters per token, takes 262,144 tokens of context and previews the Qwen4 architecture with sparse attention.#Qwen#AI Models#Open Weights
AI NewsSakana AI Signs Japan Defense Deal to Study AISakana AI signed a contract with Japan's Ministry of Defense on July 29, 2026 to study and demonstrate AI for intelligence analysis. Its value is undisclosed.#Sakana AI#Japan#AI Agents
AI NewsNVIDIA Vera CPU Specs and Why SpaceXAI Picked ItNVIDIA Vera CPU has 88 Olympus cores and up to 1.2TB/s of bandwidth, and NVIDIA claims up to 1.8x faster task completion than x86. SpaceXAI will deploy it.#NVIDIA#Vera CPU#SpaceXAI
AI NewsStability AI Funding: $76M From the Music MajorsStability AI funding round: a $76M Series B from Sony, Universal, Warner and Electronic Arts. Total funding under CEO Akkaraju is $232M, this round included.#Stability AI#Funding#Music AI
AI NewsAI Chip Smuggling: Taiwan Indicts Nine Over B300sTaiwan indicted nine over alleged B300 server smuggling. Prosecutors say 74 of 130 servers reached China via Indonesia, Japan and Hong Kong; customs held 56.#Taiwan#Export Controls#AI Servers
AI NewsUnsloth GGUF Dynamic 3.0: Where Small Quants BreakUnsloth GGUF Dynamic 3.0 quants claim over 10% better top-1 accuracy at the same size, and the smallest quant sizes stop working for agents.#Unsloth#Quantization#Local LLM
AI NewsWan 3.0 API: 30-Second Video from $0.05 a SecondAlibaba's Wan 3.0 makes up to 30 seconds of video in one pass and reads PPT, PDF and spreadsheet files as input. API pricing runs $0.05 to $0.20 per second.#Wan 3.0#Alibaba#Video Generation
AI NewsRecursive self-improvement? 153 AI research runsPrime Intellect tested recursive self-improvement claims: 153 autonomous runs on the nanoGPT speedrun, 18 models. The best closed 81.7% of the human gap.#Prime Intellect#AI agents#benchmark
AI NewsDeepMind EVE Online: What the AI Will LearnGoogle DeepMind and Fenris Creations name four research targets in EVE Online and EVE Vanguard, from continual learning and memory to multi-agent dynamics.#DeepMind#SIMA#EVE Online
GuidesGas Turbine Shortage: AI Data Centers Wait to 2031Gas turbine shortage: a GE Vernova turbine ordered in August 2026 arrives in 2031, while US data center power demand reaches 66 GW in 2027.#Gas Turbines#Data Centers#Energy
AI NewsGemma 4 31B vs Sonnet 5: same score, 40x cheaperGemma 4 31B matched Claude Sonnet 5 at 1.34 on AlphaSense’s 245 finance questions, for about 1/40 the cost per question. How it was measured and scored.#Gemma 4#Claude#open weights
AI NewsGoogle Backstory Tool: What It Can and Can't CheckGoogle Backstory checks one image for AI generation, edits and earlier appearances. India Today uses it for a first pass, then confirms by hand.#Backstory#Google#Fact-Checking
AI NewsClaude Code vs Copilot: 39% vs 21% Use at WorkClaude Code vs Copilot in JetBrains' 2026 survey: 39% of developers use Claude Code at work, 21% use Copilot. Tools overlap, so it is not market share.#Claude Code#GitHub Copilot#JetBrains
GuidesLangfuse self hosted: license, pricing, OTelLangfuse self hosted is MIT except the ee/ directories. The OTel endpoint takes OTLP over HTTP only, and Cloud pricing starts with a free 50k-unit Hobby plan.#Langfuse#LLMOps#OpenTelemetry
AI NewsLinkedIn AI Slop Button Passes a Million UsersOver a million people have used the LinkedIn AI slop button since 30 July. LinkedIn says AI slop posts now get 40% fewer views than a few weeks ago.#LinkedIn#AI Slop#Social Media
AI NewsMCP roadmap 2026: five priority areas to plan forThe MCP roadmap published in August 2026 sets five priority areas. See what they mean for the specification, Tasks, Streamable HTTP, agent identity and SDKs.#MCP#Model Context Protocol#AI Agents
AI NewsNvidia Price Increase: AI Servers Up More Than 15%Bloomberg reports the Nvidia price increase: AI servers up more than 15% in many cases from early 2027 on memory costs. Q3 gross margin guide: 74.0%.#NVIDIA#AI Servers#Memory
AI NewsChatGPT medical advice lawsuit: what OpenAI facesThe ChatGPT medical advice lawsuit says GPT-4o steered a former pastor away from the hospital. See the eight claims, the relief sought and what is unproven.#OpenAI#ChatGPT#litigation
GuidesRoboflow Playground: compare vision AI models freeRoboflow Playground runs vision AI models on your own image, for free, with no API key. Learn how its 25 tasks, Arena ELO scores and Vision Evals work.#Roboflow#computer vision#object detection
GuidesRollTab piano autocomplete: a 125M on-device modelRollTab autocompletes piano on an iPhone using a 125M model at ~108 notes/sec: MIDI tokenization, data cleaning, DPO, Core ML INT8, and what failed.#on-device AI#Transformer#Core ML
AI NewsTiangong Ultra 8.86s at World Humanoid Robot GamesTiangong Ultra ran the 100m in 8.86 seconds at the World Humanoid Robot Games, down from 9.39 on August 22. The rules the robots ran under are below.#Humanoid Robots#Tiangong Ultra#China
AI NewsTwitch AI opt-out: default opt-in and a lawsuitTwitch AI opt-out: channels start opted in to Amazon AI training and the setting applies per channel. A streamer has sued and seeks class-action status.#Twitch#Amazon#AI Training
GuidesAntigravity Remote Control: Desktop, CLI, DaemonAntigravity Remote Control covers the desktop toggle, the CLI /remote-control command, and the agy daemon for always-on machines, plus per-OS behavior notes.#Antigravity#Google#AI Agents
AI NewsLinwei Ding Google Case: Espionage Counts TossedLinwei Ding's seven economic espionage convictions were vacated on August 20, 2026. Seven trade secret counts stand, each carrying up to ten years.#Economic Espionage#Trade Secrets#Google
GuidesHubble.md: The Open-Source Notepad Agents Can EditHubble.md is a free, open-source Markdown note-taking app a coding agent can edit with you. We cover setup, HTML views, OS support and what is unfinished.#Hubble.md#Markdown#AI Agents
Guidesmodelmap: An LLM VRAM Calculator That Maps Modelsmodelmap is an LLM VRAM calculator that also maps a model's architecture. Paste a Hugging Face model ID, pick a GPU and see weights, KV cache and activations.#modelmap#Hugging Face#LLM
AI NewsNvidia in Talks With Rebellions AI: What We KnowNvidia is in early talks with Korean chip designer Rebellions AI, Bloomberg reports. We cover funding, the $2.3bn valuation, Rebel100 specs and the Groq deal.#Nvidia#Rebellions#AI Semiconductors
AI NewsTesla Robotaxi Las Vegas: What the Permit AllowsTesla robotaxi Las Vegas plans won approval for up to 5,000 vehicles, but nobody can ride yet: the conditions, the 120-day window and the airport catch.#Tesla#Robotaxi#Autonomous Driving
AI NewsWaymo Custom Chip: 5nm ASIC for Lidar and CamerasThe Waymo custom chip is a 5nm ASIC that fuses lidar, radar and camera data ahead of the driving stack. What its 1,000 TOPS covers, and what it does not.#Waymo#Autonomous Driving#Semiconductors
AI NewsWhat Claude Academy teaches, and is it free?Claude Academy has seven collections and 293 resources, from Claude 101 to a nine-hour API course. Access is free, but only the structured data says so.#Claude#Anthropic#AI learning
AI NewsAnthropic IPO: Date, Price and What Is ConfirmedAnthropic IPO date, price and ticker are unset as of Oct. 16, 2026. Bloomberg points to late November. See what is confirmed and what is only reported.#Anthropic#IPO#AI industry
AI NewsApple Music AI labels: who must tag AI musicApple Music AI labels marked Made With AI arrive later this year. Labels and distributors must now tag AI-generated music when AI made a material portion.#Apple#AI music#AI regulation
GuidesDiffusionGemma: Google's open diffusion LLMDiffusionGemma is Google's open-weight diffusion LLM: about 1,500 tokens per second on one H100, but Gemma 4 regains the throughput lead near 32 users.#diffusion#Gemma#inference
AI NewsDon't paste the AI: the nohello of AI repliesDon't paste the AI is a one-page satire asking you to stop forwarding chatbot output as your reply. It hit 1,040 points on Hacker News. Here is the argument.#AI etiquette#communication#satire
AI NewsAn AI Video Upscaler for 4K: FLUX Video UpscaleFLUX Video Upscale, an AI video upscaler, rebuilds clips at up to 4K. As of Aug. 23, 2026, it costs $0.07 to $0.10 per megapixel-second and rejects 4K input.#FLUX#video AI#image quality
AI NewsGemma Downloads Pass 1 Billion, 100,000+ VariantsGemma downloads passed 1 billion on August 20, 2026, and over 100,000 variants exist. Uses run from orbit to a cancer pathway verified in living cells.#Gemma#Google#open models
AI NewsGitHub outage: sidecar limit, VS Code retry stormThe Aug. 17 GitHub outage ran 7 hours 47 minutes. GitHub traced it to a sidecar limit that autoscaling missed; a VS Code retry bug kept Copilot down longest.#GitHub#incident#infrastructure
AI NewsGrok Build on Every Plan: App Builder, Not the CLISince August 19, Grok Build runs on every plan, on web and mobile. It turns a chat prompt into an app with a grok.me link. The CLI is a different product.#Grok#xAI#AI coding
AI NewsNVIDIA poolside deal: $6bn license, not a buyoutNVIDIA poolside deal explained: $6bn for a Model Factory license, 109 job offers and $1bn at a $12bn valuation. Not an acquisition, per the investor letter.#NVIDIA#poolside#AI industry
GuidesChatGPT Apple Messages: iMessage, SMS, RCS on MacChatGPT Apple Messages reads and sends iMessage, SMS, and RCS on a Mac. It needs Apple Silicon and Codex or ChatGPT Work; sends need approval by default.#ChatGPT#Codex#macOS
AI NewsSign Language AI: SL2T Translates ASL on Pixel 11Sign language AI from Google DeepMind: SL2T turns American Sign Language into English text in Gboard and Live Transcribe on Pixel 11. ASL only at launch.#Google DeepMind#Accessibility#Machine Translation
AI NewsLFM2.5 DSpark and QAD: Faster Decode, Better 4-bitLFM2.5 DSpark draft models from Liquid AI add speculative decoding for up to 3.18x throughput, and QAD 4-bit GGUF files keep about 97% of BF16 accuracy.#Liquid AI#On-device AI#Small Language Models
AI NewsMistral Agentic Search: 3x on FinanceBench vs RAGMistral Agentic Search vs RAG: a five-tool loop lifts FinanceBench accuracy from 26.7% to 86%; navigation alone cuts p90 latency 39.6%. When an index wins.#Mistral AI#RAG#Enterprise Search
AI NewsOrnith 1.5 Rivals Opus 4.8 on Terminal-Bench 2.1Ornith 1.5 ships in 397B, 35B and 9B sizes under MIT. The 397B scores 86.1 on Terminal-Bench 2.1 to Opus 4.8's 85.0 but trails on 8 of 13 rows.#Open Source LLM#Coding Model#Agentic AI
AI NewsNew Namazu in Sakana Translate: What 160 Tests SaySakana Translate now runs a new Sakana Namazu model and is still free. Sakana AI says it won over half of 160 in-house Japanese-to-English tests.#Sakana AI#AI Translation#Japanese LLM
AI NewsSpaceX Cognition Report: Scott Wu Denies TalksSpaceX Cognition report: Bloomberg says SpaceX tried to buy Cognition AI, maker of Devin AI. Scott Wu denied talks; a possible compute deal went unaddressed.#Cognition#SpaceX#M&A
AI NewsCerebras CS-4 claims 30x GPU inference speedCerebras CS-4, the AI accelerator announced on August 18, 2026, claims 30x GPU inference speed. See what the number measures and what is still unsaid.#Cerebras#AI chips#AI inference
AI NewsOperation Blue Skies: Atlantic contrail avoidanceOperation Blue Skies, announced August 18, 2026, uses Google AI to steer flights around contrails over Shanwick airspace. Who funds it, what changes, when.#Google#AI applications#Climate
AI NewsAgentic memory: IBM finds a dose, not a switchAgentic memory is a dose, not a switch: IBM Research tested eight models on AppWorld, and curated retrieval gave gpt-oss-120b +16.1 points for +5% tokens.#AI agents#IBM#Benchmarks
GuidesLM Studio Skills in Bionic: Use, Create, InstallLM Studio skills in Bionic use the SKILL.md format. Type @ to call one, ask Bionic to write or install one, and turn on your Claude Code or Codex skills.#Local LLM#AI agents#LM Studio
AI NewsPew AI survey: under-30s more wary than excitedA Pew AI survey finds 55% of U.S. adults under 30 more concerned than excited, up from 31% in 2021. In all, 71% expect AI to mean fewer U.S. jobs in 20 years.#AI and society#Survey data#AI and jobs
AI NewsSamsung Claude Code: chip verification in 2 daysSamsung Claude Code: System LSI reportedly cut an SoC verification job from over a month to two days, but Claude Code errors kept engineers reviewing.#Claude Code#AI case studies#Semiconductors
AI NewsAmazon rare books cut up for AI trainingAmazon rare books: 404 Media tracked one to Amazon's VGT3 team in Las Vegas, where staff cut bindings off to scan books. What model collapse research says.#Amazon#Training data#Copyright
AI NewsAnthropic Revenue Run Rate Tops $65B Ahead of IPOAnthropic told investors its Q3 2026 revenue hit $16.8 billion, about 45% above Q2, reports say. See how it fits with the $65B run rate and a November IPO.#Anthropic#IPO#AI industry
AI NewsChatGPT for Teens: Age Prediction and Study ModeChatGPT for Teens switches on from a stated age, verified age, or age prediction. See who it covers, what Study Mode does, and what parents cannot see.#ChatGPT#OpenAI#AI safety
AI NewsCursor Origin: Code Hosting Now in Early BetaCursor Origin is a code hosting platform in early beta since August 17. Here is what ships, how GitHub sync and pull requests work, and which plans get it.#Cursor#GitHub#Developer tools
AI NewsGoogle Spirit Airlines Data: A $10M Auction WinGoogle bid $10 million for Spirit Airlines’ emails and chats at a bankruptcy auction. See what the data includes, how identities are handled and who objected.#Google#Training data#Privacy
AI NewsLLM Poisoning Case: A Reported Fake Think TankLLM poisoning case: Responsible Statecraft reports 100+ Hanover Institute pieces since August 6; GPTZero flagged 11 of 12 sampled with high confidence.#AI search#Information operations#Generative AI
AI NewsAirtel Perplexity Pro: What the Free Year LeftAirtel’s free Perplexity Pro year is ending in India. Sensor Tower and Appfigures estimates track downloads, active users, and revenue after the offer closed.#Perplexity#India#Subscriptions
AI NewsPORTS-Pike: OpenAI's 8 GW Pike County Data CenterThe Pike County data center at PORTS-Pike gives OpenAI about 8 gigawatts-IT on a 20-year lease backed by NVIDIA. The first 800 MW is expected in 2028.#NVIDIA#OpenAI#Data centers
AI NewsAnthropic Model 2 Tops Mythos 5 but Stays InternalAnthropic Model 2 is somewhat more capable than Mythos 5 and has no external release plans. Here's what the August 2026 risk report says and what it redacts.#Anthropic#AI Safety#Frontier Models
GuidesData Center Economy: What Quincy Actually GotData center economy in Quincy, Washington: poverty fell from 29.4% to 6.2% and taxes funded public buildings, but an advocate says farmworkers missed out.#Data Centers#Economy#Infrastructure
AI NewsLicense Plate Reader Warrant: EFF's Line on FlockFlock license plate reader searches should need a judge-signed warrant, EFF says. What Flock Safety changed in August 2026 and what EFF says is still missing.#Surveillance#Privacy#Policy
AI NewsGemini Watermark: What the New Toggle RemovesThe Gemini watermark toggle turns off the visible mark only. SynthID and C2PA metadata stay in every file, so the output is still detectable as AI-generated.#Gemini#Google#Watermarking
AI NewsOpenAI Preparedness Team: What the FT ReportedOpenAI denies an FT report that it disbanded its Preparedness team. Here is what each side says, who runs the work now, and what changed for Dylan Scandinaro.#OpenAI#AI Safety#Reorganization
AI NewsAI Backlash: Amodei Calls It a Crisis of TrustAI backlash is a crisis of trust, not a messaging problem, Anthropic CEO Dario Amodei told Gavin Baker. Here is his case and how Zuckerberg's essay differs.#AI Policy#Anthropic#Meta
AI NewsMulti agent AI sabotage in Anthropic's testMulti agent AI sabotage in Anthropic research: three copies of one model got conflicting jobs. See how it escalated, and what multi agent systems need next.#AI Agents#Anthropic#AI Safety
AI NewsQwen Downloads: Checking the 3 Billion FigureQwen downloads passed 3 billion in six months, Bloomberg reports. Hugging Face's own report counts 2,045M on its Hub, a different window and scope.#Qwen#Open Models#Alibaba
AI NewsSpaceX Cursor Acquisition: Reading the Closing 8-KSpaceX Cursor acquisition closed August 14, 2026. The 8-K sets out 389,289,254 Class A shares for Anysphere holders, based on a $60.0 billion implied value.#Cursor#SpaceX#M&A
AI NewsStripe OpenRouter Acquisition Confirmed, No PriceStripe agreed to acquire OpenRouter on August 19, 2026 and disclosed no price. What both confirmed, why an AI gateway fits Stripe, and what stays open.#OpenRouter#Stripe#M&A
AI NewsRatepayer Protection Act Fails Senate Vote 57-43The House passed H.R. 9340 417-3, but the Senate fell three votes short of 60 on September 30. What the Ratepayer Protection Act asks of data centers.#AI Regulation#Data Centers#Energy
AI NewsOverreliance on AI cut accuracy from 27.5% to 9.2%An AI study of 3,132 people found overreliance on AI cut accuracy from 27.5% to 9.2% as confidence rose 2.5x. Paying for correct answers helped only partly.#AI Literacy#Research#Cognitive Bias
GuidesClaude Text Watermark: What a Hit Actually ProvesA Claude text watermark hit proves involvement, not authorship. Where the mark thins out, why AI text detection cannot lean on it, and who can check one.#Claude#Anthropic#Watermarking
AI NewsGoogle HEIR: A Homomorphic Encryption CompilerGoogle HEIR is an open-source compiler that turns annotated Python and pre-trained AI models into encrypted programs. See its targets, demos and cost.#Google#Homomorphic Encryption#Privacy
AI NewsNVIDIA Compute Financing: $500B Aim, Terms PendingNVIDIA compute financing: six firms aim to mobilize over $500 billion of third-party capital. What the release says, and what it leaves out.#NVIDIA#AI Infrastructure#Financing
AI NewsAI Copyright Lawsuits: Perplexity, Reddit, BraveWhere AI copyright lawsuits stand in August 2026: Dow Jones and CNN v. Perplexity, Reddit's DMCA 1201 case, and Brave v. News Corp, from dockets and reports.#Copyright#Litigation#AI Search
AI NewsQwen3.8-27B VRAM: 7GB to 56GB by Quant LevelQwen3.8-27B VRAM is 16-19GB at 4-bit and 56GB uncompressed. What each quantization level costs in memory, what long context adds, and what Apache 2.0 allows.#Qwen3.8-27B#Alibaba#Open-Weight Models
AI NewsOpen Weight Models in 2026: Qwen and Chinese LabsOpen weight models in 2026: Chinese AI labs publish the largest, Qwen has 151,448 derivatives, and sub-1B models take 83% of all-time Hugging Face downloads.#Open-Weight Models#Hugging Face#Qwen
GuidesThermodynamic Computing: Noise as the Power SourceThermodynamic computing uses thermal noise as its power source. What Berkeley Lab shows, what Extropic’s TSU chips claim, and the energy efficiency case.#Thermodynamic Computing#Semiconductors#Energy
AI NewsAI Streamer Faces: Too Perfect Loses EngagementA Journal of Business Research study of 347 AI streamers ties asymmetric faces to more engagement and typical faces or intense expressions to less.#AI in practice#Live commerce#Research
GuidesChatGPT Computer History: What It Records for YouChatGPT Computer History is off by default on macOS. It records clicks and typing, not screenshots or audio, and you can pause it or delete what it stored.#ChatGPT#OpenAI#Privacy
AI NewsClaude Code Auto Mode Is Now the Default SettingClaude Code auto mode is now the default on Pro, Max and Team. Learn what the classifier blocks, how to switch modes, and how to disable it.#Claude Code#Anthropic#Development environment
AI NewsDeepSeek Harness install guide vs Claude CodeDeepSeek Harness (dsh) is DeepSeek AI's MIT-licensed, open-source agent harness. It covers how to run it and how it compares with Claude Code and Codex.#DeepSeek#Agent#Open source
AI NewsGemini 3.7 Flash: Benchmarks and Half PricingGemini 3.7 Flash scores 65.3% on DeepSWE, up from 49.0% for 3.6 Flash. Introductory pricing is $0.75 input and $3.75 output per 1M tokens through 2026.#Gemini#Google#LLM update
AI NewsGLM 5.3 benchmarks, pricing, and open weightsGLM 5.3 keeps the GLM 5.2 base and changes only post-training. Benchmarks against Opus 4.8 and Kimi K3, the points-based Coding Plan, and the open weights.#GLM#Z.ai#Open weight
AI NewsGPT-5.6 Sol Ultrafast: Up to 14x Faster OutputGPT-5.6 Sol Ultrafast outputs up to 750 tokens per second on Cerebras, up to 14x faster than Standard. A limited OpenAI API preview since August 13, 2026.#OpenAI#GPT-5.6#GPT-5.6 Sol
GuidesMiniMax Music 3: Local Setup, VRAM and LicenseMiniMax Music 3 makes songs up to five minutes long from lyrics and a description. It fits under 24GB of VRAM, or 8GB with streaming, under a custom license.#MiniMax#Music generation#Open weight
AI NewsX Algorithm Open Source: Weights, Filters, LabelsThe X algorithm open source release on GitHub: For You feed code under Apache-2.0, what the weights scale, how filters work, and what stays unpublished.#X#Algorithm#Open source
AI NewsAI Referral Traffic Grew 8x: Shopify Q1 2026 DataAI referral traffic grew 8x and AI-referred orders 13x in Shopify's Q1 2026 data. On product pages they converted nearly 50% better; organic still leads.#AI Search#SEO#Ecommerce
AI NewsCloudflare OS: An Open-Source AI Agent PlatformCloudflare OS is the AI agent platform Cloudflare runs internally, now open source under Apache-2.0. See how Gatekeepers hold the keys and how to deploy it.#Cloudflare#AI Agents#Open Source
AI NewsDemis Hassabis Becomes Chair of Google DeepMindDemis Hassabis becomes Chair of Google DeepMind and Chief Scientist of Alphabet. Koray Kavukcuoglu steps up as SVP, and Jeff Dean is leaving Google.#Google DeepMind#Gemini#AI Industry
AI NewsJeff Dean Leaves Google for Discovery Loop StartupJeff Dean left Google after 27 years to found Discovery Loop, a startup that aims to automate the experimental loop. Google is a founding investor.#Google#AI Industry#Startups
AI NewsFLUX 3 Video: 20-Second Clips With Native AudioFLUX 3 Video makes clips up to 20 seconds long with audio generated alongside. Rates run from $0.06 a second in Draft mode to $0.95 for 4K continuation.#Video Generation AI#FLUX#Black Forest Labs
AI NewsGoogle Assistant going away: phones, not speakersIs Google Assistant going away? Google says it ends on phones and tablets from September 4, 2026. See which devices switch to Gemini and what to check.#Google Assistant#Gemini#Android
AI NewsLiquid AI LFM2.5-2.6B: Fast Agents, Not for CodingLiquid AI LFM2.5-2.6B is a small model for local agents. It decodes 220 tokens/s on an M5 Max in under 2.5GB. Liquid AI says not to use it for coding.#On-Device AI#AI Agents#Small Models
AI NewsMaple Preview: DeepGrove's Ternary Local LLMMaple Preview is DeepGrove's ternary local LLM: 5.31 GB, MIT license, and 218 tokens/sec on a Mac mini M4 via a separate runtime. Agentic results may lag.#On-Device AI#Reasoning Models#Open Source
AI NewsMuse Code: Meta's Coding Agent and Muse Spark 1.2Muse Code is Meta's terminal coding agent, in beta since August 5, 2026. How its event log resumes after a crash, what Muse Spark 1.2 adds and how to install.#Meta#Coding AI#AI Agents
AI NewsSeedance 2.5 by ByteDance: 30 Seconds in One PassSeedance 2.5 from ByteDance, released July 31, 2026: 30-second video per pass, 30 reference images, timestamp editing. See access and what the blog omits.#ByteDance#Video Generation AI#Seedance
GuidesAI Distillation Attacks in PLA-Linked ResearchWhat AI distillation is, and why a China Brief analysis of PLA-linked research on distillation attacks, watermark removal and detection evasion matters.#Knowledge Distillation#AI Regulation#Chinese AI
AI NewsCollatz Conjecture Proof: Why Lean Accepted a FakeAn AI-assisted Collatz conjecture proof passed Lean, yet the "disproof" used a kernel bug. Here is how it also fooled nanoda, and what Lean changed.#Lean#Proof Assistant#AI and Mathematics
AI NewsGoogle Earth AI Pulls Nano Banana 2 After One DayGoogle Earth AI image generation with Nano Banana 2 launched on July 30, 2026 and was rolled back on July 31 after policy-violating screenshots were shared.#Google#Image Generation AI#Nano Banana
AI NewsMiniMax H3 Open Weights: 768p Only, US/EU ExcludedMiniMax H3's open weights run locally only at 768p, because the 2K step stays on the API. The license excludes the US, EU, UK, and Korea but covers Japan.#MiniMax#Video Generation AI#Open-Weight Model
AI NewsKimi K3 Distillation: What the White House ClaimsThe Kimi K3 distillation claim: the White House says Moonshot AI distilled Anthropic's Fable. What Kratsios posted, what Moonshot denies, what is unproven.#Moonshot AI#AI Regulation#Knowledge Distillation
AI NewsOpenAI Astra: ten math proofs, no release dateOpenAI Astra, an internal version of its next major model, resolved or advanced ten open math problems. Lean 4 proofs are on GitHub; no release date is named.#OpenAI#AI and Mathematics#Lean
GuidesSakana Namazu API: pricing, setup and EU limitsSakana Namazu is Sakana AI's Japanese-tuned build of Kimi K2.6: per-token API pricing, OpenAI-compatible setup, and why it is not offered in the EU or UK.#Sakana AI#Japanese LLM#LLM API
AI NewsSolopreneurs Over $10M Nearly Tripled, Stripe SaysStripe Economics says solopreneurs passing $10M in revenue nearly tripled from 2023 to 2025 on its proxy index. It stops short of proving AI caused it.#Solopreneur#AI Adoption#Entrepreneurship
AI NewsAI cost overrun: Amazon’s 860% and AI token costsAI cost overrun at Amazon reportedly hit 860% over budget. How AI token costs make every failed retry expensive, and where AI cost control starts.#AI cost#AI agents#token billing
AI NewsAI Kill Switch Act: DHS Shutdown Power and ScopeAI Kill Switch Act, introduced July 23, 2026: who it covers, the shutdown capability required, DHS emergency orders and penalties up to $20M a day.#AI Regulation#AI Safety#United States
AI NewsAI unicorns: half of 317 have no qualifying papersAI unicorns, all 317 studied from 1998 to 2025: 52.4% had no qualifying papers and the top 10% hold 96.8% of citations. What the preprint counted.#AI Research#Academic Papers#Startups
AI NewsARC AGI 3: Two Settings Tripled OpenAI's ScoreARC AGI 3 scores for GPT-5.6 Sol rose from 13.3% to 38.3% when OpenAI kept the model's reasoning and enabled compaction instead of rolling truncation.#OpenAI#GPT-5.6#Benchmarks
AI NewsCalifornia AI Transparency Act: SB 942 and AB 853The California AI Transparency Act took effect August 2, 2026, and SB 1000 rewrote it on September 30: who is covered, three duties and the $5,000 penalty.#AI Regulation#US State Law#AI-generated Content
AI NewsContext Engineering: Claude Code's Prompt Cut 80%+Context engineering for Claude 5: Anthropic removed over 80% of Claude Code's system prompt with no measurable loss on its coding evaluations. What to trim.#context engineering#Claude Code#prompt design
AI NewsDeepSeek V4 Flash 0731: Benchmarks and MIT LicenseDeepSeek V4 Flash 0731 ships under MIT and claims wins over V4 Pro on nine benchmarks. Read the scores, the caveats, and how to call it through vLLM.#DeepSeek#open weights#AI coding
AI NewsEU AI Act 2026: What Applies in August, What WaitsEU AI Act 2026: from 2 August, chatbot and deepfake disclosure duties and AI Office enforcement apply, while high-risk rules wait until 2027 and 2028.#EU AI Act#AI Regulation#Compliance
AI NewsEU AI Act Enforcement Begins: Article 50 LabelsEU AI Act enforcement began 2 August 2026, with Article 50 labeling rules live and about 190 code of practice signatories. See what signing changes for you.#EU AI Act#AI Regulation#AI-generated Content
AI NewsGCC AI Policy: The 15-Line Rule Is Not in the TextGCC AI policy declines legally significant LLM-generated code. The text never says 15 lines; that figure comes from GNU copyright guidance.#GCC#open source#AI coding
AI NewsGemini Robotics 2: Whole-Body and Multi-RobotGemini Robotics 2 is DeepMind's VLA for whole-body robot control, announced July 30, 2026, with ER 2 and On-Device 2. Only ER 2 is in AI Studio at launch.#Gemini#Google DeepMind#robotics
AI NewsGoogle AI Studio App Canceled, Folded Into GeminiGoogle canceled the AI Studio mobile app, announced July 31, 2026, despite about 800,000 pre-orders. App building moves into Gemini; the web version stays.#Google AI Studio#Gemini#AI Development Tools
AI NewsInkling Small: Open Weights Multimodal AI ModelInkling Small is an open weights model from Thinking Machines Lab. It beats Inkling on reasoning and agentic coding, and NVFP4 runs from 180 GB of VRAM.#Thinking Machines Lab#Open-Weights Models#Multimodal
AI NewsGoogle Lyria 3.5: What Changed in Flow MusicGoogle Lyria 3.5 arrived in Flow Music. See the four upgrades, what the editing and cover tools do, and the SynthID watermark Google puts on its tracks.#Music Generation AI#Google#Lyria
AI NewsMCP Specification 2026-07-28: Sessions RemovedMCP specification 2026-07-28 makes the protocol stateless, dropping sessions and the initialize handshake. What breaks, what replaces it, and what to rewrite.#MCP#Model Context Protocol#AI agents
AI NewsChatGPT for Academic Researchers: Free for 100,000ChatGPT for Academic Researchers will bring free GPT-5.6, deep research and Codex to 100,000 researchers. Who qualifies, what's included and how to apply.#OpenAI#ChatGPT#research support
AI NewsSynthID Watermark on ChatGPT Audio: What It ProvesOpenAI added a SynthID watermark to supported ChatGPT and API audio on July 31, 2026. The verification tool now accepts audio, but a miss is no proof.#OpenAI#ChatGPT#AI-generated Content
GuidesOpenDerm: Open-source skin imaging under $8,500OpenDerm is an open-source robotic skin imaging system that captures 78 pixels per millimeter for under $8,500. It is a research tool, not a medical device.#open source#robotics#medical AI
AI NewsPharmacy AI for Refills: Kinney Drugs Pulls BackKinney Drugs added pharmacy AI to its refill line in May 2026. Delays, wrong doses and privacy worries followed; by August 10 the touch-tone line was back.#AI deployment#healthcare AI#privacy
AI NewsAI Minecraft World Generator by Sakana AISakana AI's Dream-Cubed, an AI Minecraft world generator trained on 30 billion blocks across 15 biomes: how it works and how players rated its worlds.#Sakana AI#Generative AI#Minecraft
GuidesSakana Fugu in Claude Code: Setup, Models, v1.1Run Sakana Fugu in Claude Code: one-line setup on macOS and Ubuntu, environment variables on Windows, model mapping, and what Fugu Ultra v1.1 changed.#Sakana Fugu#Claude Code#multi-agent
AI NewsZoox Robotaxi Cleared With No Steering WheelNHTSA exempted the Zoox robotaxi from parts of eight federal safety standards through July 2028, for up to 2,500 vehicles per 12 months, with conditions.#Autonomous Driving#Robotaxi#AI Regulation
GuidesBuild Apps With AI: Testing a Wine Tracker PromptCan you build apps with AI? We gave Claude Sonnet 5 one prompt for a wine cellar tracker. The first file ran unedited, but at phone width words broke apart.#Generative AI#Wine Inventory#Build Your Own
AI NewsAI Written Articles Under a Real Writer's BylineAI written articles kept appearing under a freelancer's byline after he left Clickout Media, Press Gazette reported. His GDPR claim and the 28 later articles.#AI Ethics#AI Writing#SEO
AI NewsAlphabet Q2 Earnings 2026: Cloud +82%, Search +17%Alphabet Q2 earnings 2026: revenue rose 24% to $119.8 billion, Cloud grew 82% and Search 17%. Most of the 298% net income jump came from unrealized gains.#Google#Alphabet#Earnings
AI NewsApple CXMT plan: Lutnick says no to Chinese memoryThe Apple CXMT plan would put CXMT and YMTC memory in products sold outside the US. Micron and senators oppose it, as does the administration, per Lutnick.#Semiconductors#Apple#Memory
AI NewsChatGPT Health Arrives in the US With Apple HealthChatGPT Health is rolling out to U.S. users 18 and older on web and iOS. It asks before using your linked health data and won't use it for training or ads.#ChatGPT#OpenAI#Health
AI NewsChina AI Chips: Ding Xuexiang's Reported WarningChina AI chips: sources told the WSJ that Ding Xuexiang called resisters of local chips traitors; his April 2026 self-reliance remarks are on record.#AI Chips#China#AI Regulation
AI NewsClaude Opus 5: Near-Fable Output at Half the PriceClaude Opus 5, released July 24, 2026, comes close to Fable 5 at half the price, at Opus 4.8's $5 / $25 per million tokens. Benchmarks, effort and safety.#Claude#Anthropic#LLM Updates
AI NewsGeekbench 7 vs 6: What Changed, Why Scores DifferGeekbench 7 launched July 23, 2026, with AV1, Opus and Whisper tests, a rebuilt multi-core score and CUDA support. Geekbench 6 scores don't carry over.#Geekbench#Benchmark#AI
GuidesOffline Translator App: 61 Languages, No SignalAn offline translator app that handles text, images and PDFs on your phone: the 61 languages it lists, how it works with no connection, and its limits.#Translation#Android#Offline AI
GuidesQwen TTS: Audio 3.0 Flash vs Plus, API PricingQwen TTS, shipped as Qwen-Audio-3.0-TTS, comes in Flash and Plus across 16 languages. What its benchmarks mean, how cloning works, and what the API costs.#Qwen#Text to Speech#Alibaba
AI NewsSamsung Smart Glasses Release Date and FeaturesSamsung smart glasses run Gemini, launch in fall 2026 in select markets and last up to nine hours a charge. Price and countries aren't announced yet.#Samsung#Gemini#Smart Glasses
AI NewsWaymo Uber Exclusivity to End in Austin, AtlantaWaymo Uber exclusivity in Austin and Atlanta ends in early 2028, with a Waymo app planned. It stays on Uber at least through May; split talk is an FT report.#Autonomous Driving#Waymo#AI Applications
AI NewsAMD Anthropic Deal: Up to 2GW MI450 and $5B StakeAMD Anthropic deal explained: up to 2GW of Instinct MI450 Series GPUs in Helios racks, an equity stake of up to $5B, and a Claude-ROCm tie-up.#AMD#Anthropic#AI infrastructure
AI NewsJacobian conjecture disproved with Claude Fable 5The Jacobian conjecture is false in 3+ dimensions. Levent Alpöge posted a counterexample found with Claude Fable 5; the two-variable case is still open.#Claude Fable 5#Anthropic#Jacobian conjecture
AI NewsGemini 3.6 Flash Pricing, Benchmarks vs 3.5 FlashGemini 3.6 Flash costs $1.50 input and $7.50 output per million tokens, with 49% on DeepSWE against 37% for 3.5 Flash. Compare it with Flash-Lite and Cyber.#Gemini#Google#LLM update
AI NewsAI Video Detector: NVIDIA Says Up to 92% AccurateAI video detector from NVIDIA reaches up to 92% accuracy uncompressed and 82% at 50% compression, and judges 1080p in as little as 22 ms on RTX systems.#NVIDIA#Deepfake#AI detection
AI NewsPoolside Laguna S 2.1: Open-Weight Coding ModelLaguna S 2.1 is Poolside's open-weight coding model, a 118B MoE under OpenMDW-1.1. We cover its benchmarks, how Poolside ran them, hardware, and limits.#Poolside#Open-weight#Coding AI
AI NewsChinese Open Source AI Faces Possible US CurbsChinese open source AI models may face new US restrictions. The trigger, Kimi K3, the measures under discussion, and the split inside the administration.#US-China AI#Open-source AI#Kimi K3
GuidesClaude models compared: Fable, Opus, Sonnet, HaikuClaude models as of September 2026: how Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 4.5 differ on API price, context window and plans, and which to start with.#Claude#Anthropic#Claude Fable
AI NewsCodex context window cut to 272K: why that numberOpenAI cut the Codex context window from 372K to 272K on July 18, 2026. Above 272K, GPT-5.6 bills the whole request at 2x input. Sessions now compact sooner.#Codex#OpenAI#Context window
AI NewsQwen 3.8 Open Weights: What You Actually GetQwen 3.8 open weights are text-only, always think first, and run 262,144 tokens natively. See what the hosted Qwen3.8-Max has that the download lacks.#Qwen3.8#Alibaba#Open-weight model
AI NewsAlibaba Bans Claude Code Over Hidden Tracking CodeAlibaba bans Claude Code from July 10, citing back-door risks tied to hidden code that could identify China users. Staff are told to switch to Qoder.#Claude Code#Anthropic#Alibaba
AI NewsClaude Code Token Usage: 33K vs OpenCode's 7KClaude Code token usage starts near 33,000 before your prompt, 4.7x OpenCode. CLAUDE.md and MCP servers push it to 75,000, but requests set the total.#Claude Code#Anthropic#Tokens
AI NewsKimi K3 benchmarks, pricing and open weightsKimi K3 is Moonshot AI's 2.8T-parameter open-weight model. Here is what its benchmark scores mean, what the API costs, the license terms and hardware needs.#Kimi K3#Moonshot AI#Open-weight model
AI NewsMuse Image Instagram AI Feature Removed in 3 DaysMuse Image's Instagram AI @-mention feature launched July 7, 2026 and was removed July 10. Public accounts were opted in by default; Muse Image remains.#Meta#Instagram#Image Generation AI
AI NewsSakana AI NVIDIA Partnership: Nemotron Joins FuguSakana AI NVIDIA partnership: Nemotron open models join Sakana Fugu as specialized agents. See what NVIDIA supports, how Fugu works, and what remains unsaid.#Sakana AI#NVIDIA#Open model
AI NewsApple OpenAI Lawsuit: Trial Set for July 19, 2027The Apple OpenAI lawsuit over trade secrets is set for trial on July 19, 2027. An Oct. 27 order widened Apple's early discovery. No ruling on the claims yet.#OpenAI#Apple#Lawsuit
GuidesGLM 5.2 local: run 744B on a 24 GB PC, no GPUGLM 5.2 local runs on 24 GB of RAM, not the 223 GB the buying guides quote. Why Ollama only offers a cloud tag, what the requirements say, and real speed.#GLM-5.2#Colibri#local LLM
AI NewsGrok 4.5 benchmarks, pricing, and where to use itGrok 4.5 is SpaceXAI's coding model, trained with Cursor. It leads SWE Marathon, trails GPT-5.5 on DeepSWE, and costs $2/$6 per million tokens via the API.#Grok#xAI#AI Model
AI NewsPicbreeder Rebuilt With a VLM: Can AI Be Creative?Picbreeder was a goal-free evolutionary art website. Sakana AI, MIT, and NYU replicated it with VLM agents and found where human creativity still leads.#Sakana AI#AI Research#VLM
GuidesAI Comparison: ChatGPT, Claude, Gemini PricingAI comparison for 2026: ChatGPT, Claude, Gemini, and Copilot pricing and free-plan limits, plus a use-case guide to picking the plan that fits your work.#Generative AI#ChatGPT#Claude
AI NewsIs Claude Fable 5 Back? Yes, With New SafeguardsClaude Fable 5 back worldwide since July 1, 2026: US export controls ended June 30, and a new classifier blocks the jailbreak Amazon found.#Claude#Anthropic#Fable 5
GuidesClaude Science: An AI Workbench for ResearchersClaude Science is Anthropic's beta research app for macOS and Linux. See what it does, how the reviewer agent works, who can use it, and where to get it.#Claude#Anthropic#Claude Science
AI NewsClaude Sonnet 5 vs Opus 4.8: Pricing, BenchmarksClaude Sonnet 5 launched June 30, 2026 at $2/$10 per million tokens, made permanent on August 10. It trails Opus 4.8 on five of six benchmark scores.#Claude#Anthropic#Claude Sonnet 5
GuidesEU AI Act Omnibus: New Deadlines and Nudifier BanThe EU AI Act omnibus is law: high-risk rules move to December 2027 and August 2028, and a nudifier ban and watermarking deadline hit 2 December 2026.#EU AI Act#AI Regulation#High-Risk AI
AI NewsGemini Spark for Mac: Beta, Download, Release DateGemini Spark for Mac is in beta for US Google AI Ultra subscribers aged 18+. What it does with files, how to download it, connected apps, MCP, rollout dates.#Google#Gemini#AI Agent
AI NewsNano Banana 2 Lite: 4 Seconds, $0.034 an ImageNano Banana 2 Lite makes an image in about 4 seconds for $0.034 per 1K image. Pricing, comparison with Nano Banana 2 and Pro, and Gemini Omni Flash video.#Google#Gemini#Image Generation AI
AI NewsLongCat 2.0: Meituan's 1.6T Open MoE Coding ModelLongCat 2.0 is Meituan's open 1.6T MoE coding model with a 1M-token context. Covers the domestic-chip training claim, SWE-bench Pro 59.5 and how to run it.#Meituan#LongCat#Open Source AI
AI NewsProton Lumo 2.0: Privacy Design and What ChangedProton Lumo 2.0 is a privacy-first AI assistant under Swiss law. Proton reports an index score of 51, up from 15, plus reasoning modes and image tools.#Proton#Lumo#Privacy
AI NewsAI Agent Model at 35B Rivals Trillion-Parameter AIAn AI agent model at 35B, Agents-A1 from Shanghai AI Lab, claims trillion-parameter performance. Its MoE weights are public under Apache 2.0.#Agents-A1#Shanghai AI Lab#AI Agent
GuidesX MCP Server: Give Claude and Grok the X APIThe official X MCP server lets Claude, Cursor and Grok call the X API with your own account permissions. Two servers, OAuth or token auth, and client setup.#X#MCP#Model Context Protocol
AI NewsClaude Mythos 5 Returns for Critical InfraClaude Mythos 5 returned for US critical-infrastructure groups after a June export-control suspension, with access still limited through Project Glasswing.#Claude#Anthropic#Mythos 5
GuidesColorado AI Act: Repealed Before It Took EffectThe Colorado AI Act (SB24-205) was repealed and replaced by SB26-189 before its effective date. Summary of the old law, the delay and the 2027 rules.#Colorado AI Act#AI regulation#high-risk AI
AI NewsOpenAI Jalapeno Chip: Inference Benchmarks 2026OpenAI Jalapeno, the Jalapeño inference chip, posted benchmarks Aug 25, 2026: 1.5 to 1.9x more work per watt at peak throughput than Nvidia GB200 and GB300.#OpenAI#Broadcom#AI chip
GuidesChatGPT Summarizer: Prompts, Splits, File LimitsChatGPT summarizer guide: prompts for bullet and word-limit summaries, splitting long documents by chapter, upload limits, and keeping private text private.#ChatGPT#OpenAI#Summarization
AI NewsGLM 5.2 benchmarks, pricing, and Claude Code setupGLM 5.2 is Z.ai's 744B-parameter open-weight model with a 1M-token context. Compare benchmarks with GPT-5.5, check API pricing, and set it up in Claude Code.#GLM#Open-weight model#Z.ai
GuidesGPT-5.6 availability by plan: Sol, Terra and LunaGPT-5.6 availability: generally available since July 9, 2026, but Sol, Terra and Luna reach different plans and tools. Here's who gets what and what it costs.#OpenAI#GPT-5.6#ChatGPT
GuidesGPT-5.6 Sol, Terra and Luna: Specs and PricingGPT-5.6 Sol, Terra and Luna are OpenAI's three tiers, generally available across ChatGPT, Codex and the API, with pricing cut twice since launch.#OpenAI#GPT-5.6#GPT-5.6 Sol
AI NewsCoffeeBench by Sakana AI: How It Works and ResultsCoffeeBench is Sakana AI's benchmark where an LLM runs a coffee roaster for 90 days, scored on net income. See how it works and how GPT-5.5 and Claude did.#Sakana AI#AI agents#Benchmark
AI NewsChatGPT Market Share Falls Below 50% in 2026ChatGPT market share slipped below half in March 2026 and fell to 46.4% by the end of May, per Sensor Tower. Gemini holds 27.7%, Claude 10.3%.#ChatGPT#AI Assistant#Market Share
GuidesChatGPT Slack Integration: Actions, Setup, LimitsChatGPT Slack integration lets ChatGPT search Slack and, if an admin allows it, take actions. See the connector's actions, setup, OAuth scopes and limits.#ChatGPT#OpenAI#Slack
AI NewsChina AI data centers: $295B plan, 80% local chipsChina AI data centers would get about $295 billion over five years, with at least 80% of chips from local suppliers. US firms top that in 2026 alone.#China AI#AI data centers#US-China AI race
GuidesClaude Tag: Anthropic's AI teammate for SlackClaude Tag puts one shared Claude in your Slack channels. See how mentions, ambient mode, and routines work, which plans include it, and how admins set it up.#Claude#Anthropic#Claude Tag
GuidesCodex Record and Replay: Mac Workflows as SkillsCodex Record and Replay turns a demo on your Mac into a reusable skill. Learn how to record and replay it, what it needs, and when a plugin fits better.#OpenAI#Codex#Workflow automation
AI NewsMistral OCR 4: Bounding Boxes, Pricing and APIMistral OCR 4 returns bounding boxes, block types and confidence scores in 170 languages. It costs $4 per 1,000 pages, $2 in Batch and $5 for Document AI.#Mistral#OCR#Document AI
GuidesApertus AI: The Swiss Open LLM for Sovereign AIApertus AI is a fully open Swiss LLM from EPFL, ETH Zurich and CSCS. See the 8B, 70B and v1.1 specs, benchmarks, license and how to run it.#Apertus#Sovereign AI#Open-source LLM
GuidesClaude Fable 5 Unavailable: The Ban and Your PlanClaude Fable 5 unavailable? A US export directive banned it on June 12, 2026; it returned July 1. Today the cause is your plan, app version, or a fallback.#Claude#Anthropic#Fable 5
GuidesClaude identity verification: passing the ID checkClaude identity verification: Persona checks a physical photo ID and sometimes a selfie. See which IDs pass, who keeps the images, and what to do if it fails.#Claude#Anthropic#Privacy
GuidesChatGPT file upload limit: which one stops youChatGPT file upload limit: 512MB per file and 80 files every 3 hours, but the hidden quota and the Enterprise 110k-token cut are what stop most people.#ChatGPT#OpenAI#PDF
AI NewsOpenAI investigation: what state AGs subpoenaedOpenAI investigation: a coalition of state attorneys general subpoenaed records on ads, consumer and health data and minors, reports said on June 12, 2026.#OpenAI#AI regulation#ChatGPT
GuidesPaddleOCR PP-OCRv6: one model, 50 languagesPaddleOCR PP-OCRv6 reads 50 languages with one model and claims to beat VLMs thousands of times larger. See whose benchmark backs it and what to test.#PaddleOCR#OCR#Text Recognition
AI NewsProject Fetch: Claude Opus 4.7 Drives a Robot DogProject Fetch Phase 2 in numbers: Claude Opus 4.7 finished robot dog tasks about 20x faster than the fastest human team, but could not fetch the beach ball.#Claude#Anthropic#Project Fetch
GuidesClaude Opus 4.6 vs 4.7 vs 4.8 pricing comparedClaude Opus 4.6, 4.7 and 4.8 all cost $5 input and $25 output per million tokens, but 4.7 and 4.8 use a tokenizer that can bill up to 35% more tokens.#Claude#Anthropic#Pricing
AI NewsSakana Fugu Ultra is not always the best FuguSakana Fugu Ultra leads 7 of 12 published benchmarks, and on three of the rest the cheaper Fugu scores higher. Prices, billed orchestration tokens, EU access.#Sakana AI#Multi-agent#LLM update
GuidesChatGPT pricing: Free, Plus, and Pro with GPT-5.6ChatGPT pricing for 2026: what Free, Go, Plus, and Pro cost, which plans include GPT-5.6 Sol, and how to choose by the usage limit you hit.#ChatGPT#OpenAI#GPT-5
GuidesWhat Is Claude AI? Free Plan, Models, vs ChatGPTWhat is Claude AI? Anthropic's AI assistant. Start free on Sonnet; paid plans add more models and higher usage. Against ChatGPT, pick it by task.#Claude#Anthropic#Claude Opus
AI NewsClaude Opus 4.8: Pricing, Fast Mode, WorkflowsClaude Opus 4.8 launched May 28, 2026 at $5 and $25 per million tokens. See what changed from Opus 4.7, plus fast mode, dynamic workflows and effort control.#Claude#Anthropic#LLM updates
GuidesGemini 3.5 Pro is still not in the Gemini APIGemini 3.5 Pro has not shipped. Google's pricing page lists Flash up to 3.8 but no Pro past 3.1. See what to call, what it costs, and when prices rise.#Gemini#Google#Gemini 3.5
GuidesGrok Pricing 2026: Free, SuperGrok, and API CostsGrok pricing right now: Free, SuperGrok, and SuperGrok Plus costs verified on x.ai, current xAI API rates per model, and what changed since Grok 4 retired.#Grok#xAI#Grok 4
AI NewsLocal LLM Coding Now Hits 75% of Cloud AILocal LLM coding now reaches 75% of frontier cloud AI on a 2022 Mac. What one detailed report shows about the models, cost, and limits.#Local LLM#Open-weight models#Ollama