Gemini 3.7 Flash: Benchmarks and Half Pricing

What Gemini 3.7 Flash is and when it launched

On August 13, 2026, Google introduced Gemini 3.7 Flash as its most intelligent workhorse model yet for coding and agents. An agent is an AI program that plans and carries out several steps on its own, such as calling tools, instead of answering one prompt at a time.

The release came three weeks after Gemini 3.6 Flash. Google says the release is a direct result of developer feedback and algorithmic innovations, and that it plans to bring those innovations to future models. The stated focus is software engineering, knowledge work and web development workflows.

The previous release is covered in Gemini 3.6 Flash Pricing, Benchmarks vs 3.5 Flash. The model belongs to the Gemini family.

Gemini 3.7 Flash benchmarks against Gemini 3.6 Flash

Gemini 3.7 Flash scores higher than 3.6 Flash on all five metrics Google published, and DeepSWE v1.1 rises from 49.0% to 65.3%. Every score below comes from Google's announcement.

DeepSWE v1.1 and AutomationBench scores (%)

DeepSWE rose 16.3 points and AutomationBench 13.4 points. AutomationBench's 3.7 Flash score is about 1.8 times its 3.6 Flash score, against about 1.3 times for DeepSWE.

0204060

65.3DeepSWE 3.7

49.0DeepSWE 3.6

30.4Automation 3.7

17.0Automation 3.6

“DeepSWE v1.1 and AutomationBench scores (%)” lists all five metrics with the gain, which is the difference between the two scores. Four gains are in percentage points, and the WebDev Arena gain is in Elo points.

Google cites FrontierCode and DeepSWE for coding performance. WebDev Arena rates web development results with Elo, a scale where a higher number means a stronger showing in head-to-head comparisons. Google describes GDP.pdf as a test of how well a model processes complex documents, and AutomationBench as a test of completing real-world business workflows.

DeepSWE shows the largest gain in points. In relative terms AutomationBench grew the most, from 17.0% to 30.4%.

What Google says improved for coding and agents

Google lists gains in debugging, first-pass accuracy, web layouts and day-to-day agent behavior. All of these are Google's own claims:

• Coding: stronger results in debugging and issue resolution, with higher first-pass code accuracy.
• Web development: more functional layouts and complete apps in fewer prompts.
• UI generation: close adherence to a reference input, such as a screenshot, an image or a full design system.
• Developer experience: it adapts to roadblocks better, asks for clarification when needed and follows instructions more faithfully.
• Planning: more effort goes into multi-step plans and tool calls.
• Knowledge work: better reasoning and accuracy in knowledge-dense fields such as finance, law and biosciences.

Google sums this up as more disciplined execution, which it says means less manual oversight and fewer retries across engineering workflows. Google has not disclosed counts for retries or oversight, so the size of the effect is unknown.

Gemini 3.7 Flash pricing and the date it doubles

Gemini 3.7 Flash costs $0.75 per 1M input tokens and $3.75 per 1M output tokens through December 31, 2026, and both rates double on January 1, 2027. A token is a small unit of text that the model reads or writes, and API prices are set per million of them.

Google calls this rate an introductory price and says it is half the original 3.6 Flash cost per million tokens. A footnote in the post sets the end date.

Gemini 3.7 Flash price per 1M tokens

Period	Input	Output
Through December 31, 2026	$0.75	$3.75
From January 1, 2027	$1.50	$7.50

A project that starts in November 2026 and keeps running into the new year will pay twice the unit price from January 1, 2027. The price table above gives both rates.

Where you can use Gemini 3.7 Flash

Google offers Gemini 3.7 Flash to developers, enterprises and individual subscribers through three separate routes. The route depends on who is using it.

• Developers: Google Antigravity, or the Gemini API through Google AI Studio and Android Studio. Google also points to a developer guide.
• Enterprises: the Gemini Enterprise Agent Platform and the Gemini Enterprise app.
• Individuals: Gemini Spark in the Gemini app, for Google AI Pro and Ultra subscribers in supported countries.

How Gemini Spark uses Gemini 3.7 Flash

Google said on August 13, 2026 that Gemini Spark would use Gemini 3.7 Flash starting that day, for Google AI Pro and Ultra subscribers in more than 160 countries. Google launched Spark at its I/O event as a personal AI agent that runs 24/7 and takes action on the user's behalf under the user's direction.

Google says the model change makes Spark more efficient for knowledge work. Tool use improves for Google Workspace apps, and accuracy and output quality improve on complex workflows that need several skills. The examples Google gives are consolidating files, drafting emails and updating status documents; the post gives no figures for the Spark change.

Safeguards shipped with Gemini 3.7 Flash

Google says Gemini 3.7 Flash ships with updated safeguards against misuse in chemical, biological, radiological and nuclear (CBRN) areas and in cyber offense. CBRN stands for those four kinds of weapons and hazards.

Google adds that the model enables beneficial use cases alongside the safeguards. Google points readers to the 3.7 Flash model card for details. The earlier Gemini 3.5 generation is covered in Gemini 3.5 Pro is still not in the Gemini API.