Gemini 3.5 Pro is still not in the Gemini API

Gemini 3.5 Pro is not a model you can call

Anyone waiting for Gemini 3.5 Pro has no model ID to call today. As of September 5, 2026, neither the Gemini API pricing page nor the Vertex AI model list contains the words "3.5 Pro" or the model ID gemini-3.5-pro.

An API request has to name a model ID. Without one there is nothing to send the request to, so a plan that depends on that model cannot be run. What Google did release from the 3.5 family is Flash, followed by Flash-Lite. Its announcement said Google was kicking off the series by releasing 3.5 Flash, and the Vertex AI page for that model gives May 19, 2026 as the release date.

The pages show an absence, and an absence is a smaller claim than a cancellation. Nothing on those pages says the model is dead, and Google's Gemini 3.6 Flash announcement says 3.5 Pro is testing with partners, with broad availability planned once it is ready.

Flash reached 3.8 while Pro stayed at 3.1

The pricing page answers the question by its own layout. It carries a section for each of Gemini 3.8 Flash, 3.7 Flash, 3.6 Flash and 3.5 Flash, and the newest Pro-class section is Gemini 3.1 Pro Preview.

A Gemini 3.5 Pro entry would sit after 3.1 Pro Preview in the Pro line, and the page has no such entry. A buyer who comes looking for it finds Flash models listed at each step from 3.5 to 3.8 instead. Gemini 3.6 Flash and 3.7 Flash are covered in "Gemini 3.6 Flash Pricing, Benchmarks vs 3.5 Flash" and "Gemini 3.7 Flash: Benchmarks and Half Pricing". Each one covers the benchmarks and pricing for that generation.

What the Gemini API pricing page lists, as of September 5, 2026

Prices are per 1M tokens on the paid tier. The Gemini 3.5 Pro row is the only one with nothing in any column, and the Gemini 3.8 Flash row is the only one whose price carries an end date.

Model	Input	Output
Gemini 3.8 Flash	$0.75 through December 31, 2026, then $1.50	$3.75 through December 31, 2026, then $7.50
Gemini 3.5 Flash	$1.50	$9.00
Gemini 3.5 Flash-Lite	$0.30	$2.50
Gemini 3.1 Pro Preview	$2.00, or $4.00 above 200k tokens	$12.00, or $18.00 above 200k tokens
Gemini 3.5 Pro	not listed	not listed

Gemini 3.1 Pro Preview is the newest Pro you can pay for

Take a job a Flash model has already failed, such as changes across a whole repository that depend on each other. The newest Pro-class model with a price on the page is Gemini 3.1 Pro Preview. Call it by its model ID, gemini-3.1-pro-preview. A second ID, gemini-3.1-pro-preview-customtools, is listed beside it.

It is Google's third-generation Pro model, built for multimodal understanding, agentic capabilities and vibe-coding. The price is $2.00 per million input tokens and $12.00 per million output tokens (as of September 2026). Against 3.5 Flash at $1.50 in and $9.00 out, the dollar gap is $0.50 on the input side and $3.00 on the output side, which is where a model that writes long answers spends its money.

Prompts over 200k tokens double the input price

A prompt of more than 200k tokens is billed at $4.00 in and $18.00 out: double the input rate and one and a half times the output rate of a shorter prompt. The step applies to the Pro model only, and none of the Flash rows listed above have it.

Reading a large repository is a job a team may turn to Pro for, and a repository passes 200k tokens quickly. The input charge for that one request then doubles and the output charge rises by half without anything about the task changing, so the larger rates are the ones to put in a budget for that kind of work.

Gemini 3.5 Flash costs $1.50 in and $9.00 out per million tokens

A token is a small piece of text, usually part of a word, and Google bills the tokens you send and the tokens the model sends back as two separate counts. For Gemini 3.5 Flash on the paid tier, a million tokens going in cost $1.50 and a million tokens coming out cost $9.00 (as of September 2026).

Each output token costs six times as much as an input token, so a request that writes a lot costs more than a request that reads the same amount. Take two jobs on the same model. Summarizing a long report into a page is the cheaper job, because most of its tokens are input. Producing a long report from a short brief costs more, because most of its tokens are output, and output is the expensive side.

Context caching, which lets the model reuse a long prompt without charging full input price again, costs $0.15 per million tokens. It also carries a second charge that is easy to miss. Cached text is billed for storage at $1.00 per million tokens per hour, so the longer a cache is kept, the more it costs.

The newer Flash models are half price only until the end of 2026

Google calls Gemini 3.8 Flash its most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents and complex enterprise workflows. At $0.75 in and $3.75 out per million tokens (as of September 2026), it costs less than 3.5 Flash on both sides right now. The input price is half of what 3.5 Flash charges, and the output price is under half.

Those two figures carry a date in the same table cell. They hold through December 31, 2026, and from January 1, 2027 the price becomes $1.50 in and $7.50 out. Anything costed on 3.8 Flash today has to be recalculated on a date Google has already published, and the new input price equals what 3.5 Flash charges now.

A budget built on the $0.75 rate will be wrong in January even if nobody changes it. Write the end date next to the price in the estimate, or work out both versions. Anyone planning a project that runs past the end of 2026 should use the January figures from the start.

Flash-Lite is the tier for volume, not for hard problems

Picture a team that has to translate a few thousand product descriptions every night. Each call is small and nobody reads the output closely, so the cost per call is what matters. Gemini 3.5 Flash-Lite charges $0.30 per million input tokens and $2.50 per million output tokens (as of September 2026). That is a fifth of the input price of 3.5 Flash and a little over a quarter of its output price.

The model is built for that job: it is cost-efficient and made for high-volume agentic tasks, translation and simple data processing. Google's description does not mention hard problems. Sorting, extracting and translating at scale fit that description, and work where a wrong answer is expensive is better sent to a larger model.

Hard cases are where a full Flash model earns its higher price. A reasonable rule is to run the bulk of a routine job on Flash-Lite, check the results, and send the items that fail the check to 3.5 Flash.

The free tier is paid for with your prompts

Take a developer who builds a prototype on the free tier and later feeds it real customer text. The pricing page has a row that matters for that case. Under "Used to improve our products", the Free Tier column reads Yes and the Paid Tier column reads No. Free usage costs nothing in money, and what the developer gives up instead is the prompts, which Google uses to improve its products.

So the free tier is a different arrangement from the paid one, and not a discounted version of it. It suits tests with public or invented text. Customer data, internal documents and anything under a confidentiality agreement belong on the paid tier, where the same row reads No.

Search grounding, which lets the model check its answer against Google Search results, is metered apart from tokens. Google gives 5,000 free search requests per month, then charges $14 per 1,000 requests (as of September 2026). The allowance is shared across all Gemini 3.x models, so running the same app on three models does not buy three allowances. A service that grounds most of its answers passes the free 5,000 at about 170 answers a day, if each answer uses one search request.

Retirement dates sit on Vertex AI model pages, not the pricing page

Suppose a company wires a model ID into a billing system that has to keep running for years. The Gemini API pricing page lists prices and has no column for how long a model will stay available, so the answer is on a second page.

Vertex AI, Google Cloud's platform for running its models, answers it on each model's page. The page for Gemini 3.5 Flash gives a launch stage of GA (generally available), a release date of May 19, 2026 and a retirement date of May 19, 2027 or later. The page for 3.5 Flash-Lite gives the same three fields with July 21 dates: released July 21, 2026, retiring July 21, 2027 or later. "Or later" means Google may keep a model past that day, so for a team that runs on Vertex AI the date works as the earliest day to plan a migration around. A team that calls the Gemini API will not find a retirement date for its model on the pricing page.

The Vertex AI page for 3.5 Flash also lists the limits a request runs into. Gemini 3.5 Flash accepts a context window of 1,048,576 tokens and returns at most 65,536 tokens in one answer, so no prompt can make a single answer longer than that.

Pick a Flash tier instead of waiting for 3.5 Pro

Start from the work in front of you and ignore the release date, because neither the pricing page nor the Vertex AI list gives a release date. A team that translates thousands of short texts a night should use Flash-Lite. A team that needs long, careful answers should use 3.8 Flash. A team whose Flash model has already failed on hard reasoning should try 3.1 Pro Preview.

The output price is what separates the two Flash models. Gemini 3.8 Flash costs less on output than 3.5 Flash both now and after the price rise: $3.75 against $9.00 now, and $7.50 against $9.00 from January 1, 2027. Input costs the same as 3.5 Flash after the rise. A project that will still be running in 2027 should therefore be costed at the January figures, which are $1.50 in and $7.50 out.

Reach for Pro only after a Flash model has failed the task. Its output costs $12.00 per million tokens, and prompts above 200k tokens raise the rates to $4.00 in and $18.00 out. For tests that use invented or public text, the free tier costs nothing, but it should never carry customer data.

Which Gemini model to call, by the work in front of you

Prices are per 1M tokens as of September 2026. Three of the five rows point to a Flash tier and one to Pro; the last row is Gemini 3.5 Pro itself.

The work	Model to call	Why
Thousands of small, routine calls such as translation or data sorting	Gemini 3.5 Flash-Lite	$0.30 in and $2.50 out, and the model is built for this kind of work
Long, careful answers	Gemini 3.8 Flash	$3.75 out through December 31, 2026, then $7.50, still below the $9.00 of 3.5 Flash
A project that runs into 2027	Gemini 3.8 Flash, costed at the January figures	$1.50 in and $7.50 out from January 1, 2027
Hard reasoning that a Flash model has failed	Gemini 3.1 Pro Preview	The newest Pro-class model with a price, at $2.00 in and $12.00 out
Any work at all	Not Gemini 3.5 Pro	No model ID is listed for it, so there is nothing to call

Use a Flash tier now instead of waiting for 3.5 Pro

Gemini 3.5 Pro has no model ID on Google's pricing page and no entry on the Vertex AI model list, so a plan that depends on it has nothing to call. The page lists Flash at 3.5, 3.6, 3.7 and 3.8, with 3.5 released on May 19, 2026, and the newest Pro-class model there is still Gemini 3.1 Pro Preview.

• Small, routine calls in large numbers: Gemini 3.5 Flash-Lite at $0.30 in.
• Long or careful answers: Gemini 3.8 Flash, costed at $1.50 in and $7.50 out for any month after December 2026.
• Hard reasoning that a Flash model has failed: Gemini 3.1 Pro Preview at $2.00 in and $12.00 out, with $4.00 in and $18.00 out above 200k tokens.
• Customer data: the paid tier, because the free tier lets Google use prompts to improve its products.

Whoever has put a project on hold for 3.5 Pro can restart it on one of these today. The cheap price on 3.8 Flash ends on December 31, 2026, so usage from January 1, 2027 is the first to be billed at the higher rate.