by Google

Gemini 3.5 Pro review, pricing and limits

The Pro tier Google announced for Gemini 3.5 in May 2026, not yet released and with no published context window, price or benchmark results.

Not for
Any workload that must run now
  • Preview
  • Proprietary

Should you use Gemini 3.5 Pro?

Gemini 3.5 Pro is not available to call. Google announced it at I/O in May 2026 and promised a rollout the next month, but the Gemini API still lists Gemini 3.1 Pro Preview as its only Pro model. Teams that need a Pro-tier Gemini today should build on 3.1 Pro Preview.

Not for

  • Any workload that must run now There is no endpoint to call; Gemini 3.1 Pro Preview and Gemini 3.8 Flash are the Google options available today.

Gemini 3.5 Pro pricing

Pricing model
per-token

Google has not set a price for Gemini 3.5 Pro. Its Gemini API pricing page, checked at the end of September 2026, has no entry for the model, and the only Pro model it prices is Gemini 3.1 Pro Preview. Per-token numbers quoted for 3.5 Pro elsewhere are estimates, not Google's.

What Gemini 3.5 Pro can do

Google has not published a context window for Gemini 3.5 Pro. The model is absent from the Gemini API model list in late September 2026, and its announcement gives no token limit. The Gemini models developers can call today document a limit of roughly one million input tokens instead.

Key features

  • Pro tier of the Gemini 3.5 family

    Presented on the I/O keynote stage as the larger sibling of Gemini 3.5 Flash, the family member Google shipped that day.

  • In internal use before release

    Google's announcement said the model already ran internally, with a public rollout planned for June 2026 that has not happened yet.

Known limitations

  • Unreleased at the end of September 2026, with no public model ID or API access.
  • Google has not published its token limits, pricing or evaluation results.

Who it is for

  • Google Cloud teams deciding whether to wait for the next Pro-tier Gemini
  • Developers checking whether a 3.5 Pro model ID exists before planning a migration

Gemini 3.5 Pro benchmarks

Standout result

132

Days between Google's I/O announcement and 28 September 2026 with no public Gemini 3.5 Pro release

Source: Google Gemini 3.5 announcement (19 May 2026) and Gemini API changelog, checked 28 September 2026

Strengths and limits

Strengths

  • Google committed to a Pro tier for the Gemini 3.5 generation in public, so the model sits on an announced roadmap rather than in rumour.
  • The family's shipped member, 3.5 Flash, is stable on the Gemini API, giving teams a documented baseline for the generation while they wait.
  • Named by Google at I/O 2026 as the Pro tier of Gemini 3.5.

Limits

  • Not released: no model ID, no API access and no new rollout date from Google as of late September 2026.
  • No published context window, output limit, price, benchmark score or model card.
  • The June 2026 rollout Google announced did not happen, so any plan pinned to a launch date is guesswork.

Gemini 3.5 Pro alternatives

ModelLabContextIn / out per 1MReads and writes
Gemini 3.5 Pro This pageGoogle------
Gemini 3.1 ProPick Gemini 3.1 Pro: it is the Pro-tier Gemini you can actually call, with a documented token limit and a published price; revisit 3.5 Pro only after Google lists it.Google1M$2 / $12text, image, audio, video, pdf → text
Gemini 3.5 FlashGemini 3.5 Flash is the half of the 3.5 family that shipped; use it for fast, lower-cost work and wait for Google's own numbers before judging the Pro tier against it.Google1M$1.50 / $9text, image, audio, video, pdf → text
GPT-5.5GPT-5.5 is available to build on now, while Gemini 3.5 Pro has no endpoint, so there is no head-to-head to run until Google releases it.OpenAI1.1M$5 / $30text, image → text

Also compared

  • Claude Opus 4.8Choose Claude Opus 4.8 for frontier work today; a fair comparison with Gemini 3.5 Pro needs Google's published limits and benchmarks, which do not exist yet.

Gemini 3.5 Pro specs

Architecture
Not disclosed
Parameters
Not disclosed

Training data

Google has not published a model card or a training data summary for Gemini 3.5 Pro.

Known quirks

Show 3 known quirks
  • No public model ID

    The Gemini API has no 3.5 Pro model page; ai.google.dev URLs for gemini-3.5-pro and gemini-3.5-pro-preview return 404 as of 28 September 2026.

    Workaround: Build against gemini-3.1-pro-preview and change the model string when Google lists 3.5 Pro.

  • Circulating specs are unofficial

    Context-window, pricing and benchmark figures attributed to 3.5 Pro come from third-party coverage, not from Google.

    Workaround: Check the Gemini API models and pricing pages before sizing inputs or setting a budget.

  • Missed rollout month

    Google's May 2026 post said 3.5 Pro would roll out the following month, and its developer documentation has named no later date since.

    Workaround: Plan on shipped models and schedule 3.5 Pro as a later evaluation rather than a launch dependency.

Gemini 3.5 Pro facts

Licence
Proprietary
Safety posture
Balanced
Released
19 May 2026
Family
Gemini 3.5
Variant
Multimodal

Version history

  • Gemini 3.5 Pro (announced) · 19 May 2026

    Announced with Gemini 3.5 Flash at Google I/O; described as in internal use with a rollout the following month. Not listed on the Gemini API as of 2026-09-28.

  • gemini-3.1-pro-preview · 01 Feb 2026

    Predecessor and the Pro model Google lists today, in preview, with 1,048,576 input and 65,536 output tokens.

What changed

  • Release Gemini 3.5 Flash launches GA at I/O; Pro remains in enterprise preview source
  • Release Gemini 3.5 Pro announced at Google I/O 2026 with 2M context and Deep Think (gemini-3.5-pro-preview) source

Questions about Gemini 3.5 Pro

What is the context window of Gemini 3.5 Pro?

There is no official figure. Google's I/O post of 19 May 2026 names no token limit for 3.5 Pro, and no Gemini API model page exists for it as of 28 September 2026, so reports of a 2M-token window remain unconfirmed. For comparison, Gemini 3.1 Pro Preview documents 1,048,576 input and 65,536 output tokens.

Has Google released Gemini 3.5 Pro yet?

No. Google said in May that 3.5 Pro was in internal use and would roll out a month later, yet the Gemini API changelog, models page and pricing page still carry no 3.5 Pro entry. The newest Flash model, Gemini 3.8 Flash, became generally available on 2 September 2026.

What will Gemini 3.5 Pro cost?

Nobody outside Google knows yet. The model has no row on the Gemini API pricing page, so any per-token figure you see is a guess. Gemini 3.1 Pro Preview is the Pro-tier model with a published price today; read its current rates on that same pricing page before budgeting.

Which Gemini model should I use while 3.5 Pro is unavailable?

Pick Gemini 3.1 Pro Preview for the hardest reasoning and coding work, and Gemini 3.8 Flash where speed and cost matter more; Google's models page recommends 3.8 Flash for new projects. You can try either in AI Studio right now, and moving to 3.5 Pro later should mean editing one model string.

Does Gemini 3.5 Pro include Deep Think?

Google has not said so. Its Gemini 3.5 announcement does not mention Deep Think, and with no 3.5 Pro model card there is no official list of reasoning modes. Treat Deep Think on this model as unconfirmed until Google publishes documentation for it.

About Gemini 3.5 Pro

Gemini 3.5 Pro is the Pro tier of the Gemini 3.5 family that Google and its Google DeepMind lab announced at Google I/O on 19 May 2026. Only Gemini 3.5 Flash shipped that day. Google's post said 3.5 Pro was "already being used internally" and would roll out the following month. As of 28 September 2026 it has not: the model is missing from the Gemini API models page, the Gemini API changelog (latest entry 22 September 2026) records no 3.5 Pro release, and Google documentation URLs for a 3.5 Pro model page return 404.

Read the full background

There is no Google-documented token limit for Gemini 3.5 Pro. The 2M-token figure that circulates in coverage of the model does not appear in Google's announcement or in any Google model documentation, so Hokai does not state it as a spec. What Google does document is the limit on the models you can call today. Their Gemini API model pages, read on 28 September 2026, give these figures:

| Model | Status on the Gemini API | Input token limit | Output token limit | |---|---|---|---| | Gemini 3.5 Pro | Not listed | Not published | Not published | | Gemini 3.1 Pro Preview | Preview | 1,048,576 | 65,536 | | Gemini 3.8 Flash | Stable, GA on 2 September 2026 | 1,048,576 | 65,536 | | Gemini 3.5 Flash | Stable, now labelled legacy | 1,048,576 | 65,536 |

Pricing, benchmarks, a model card, a model ID and a knowledge cutoff are unpublished too. The Gemini API pricing page has no 3.5 Pro row; its only priced Pro model is 3.1 Pro Preview. Deep Think, which third-party write-ups often attach to 3.5 Pro, is not mentioned in Google's Gemini 3.5 announcement.

Google's Flash line kept moving while the Pro release waited. Gemini 3.6 Flash reached general availability on 21 July 2026 and Gemini 3.8 Flash followed in September. For new projects, Google's models page now points developers to 3.8 Flash or Gemini 3.5 Flash-Lite.

Who should not wait for it: any team with a production deadline. A Pro-tier preview and several stable Flash models are callable now through Google AI Studio and the Gemini API, and outside Google, GPT-5.5 and Claude Opus 4.8 are frontier alternatives with their own Hokai profiles. The case for waiting is narrower: a team already committed to Gemini on Google Cloud, whose own evaluations show 3.1 Pro Preview falling short on hard reasoning, and which can hold a decision until Google publishes the new model's limits and price. Even then, build against a model ID that exists today and treat 3.5 Pro as a later swap to test, not a dependency. For coding work specifically, the AI coding assistants guide covers which shipped models the main tools run on.

This profile will carry Gemini 3.5 Pro's documented limits, price and benchmarks once Google publishes them; until then every figure here is either Google's own or marked as unpublished. Other Google models sit in the model directory, announced-but-unshipped entries appear under the preview status filter, the consumer app has its own Gemini profile, and the AI chatbot guide compares the assistants built on these models.

Sources and guides