Gemini 3.5 Pro
- Provider
- Status
- Preview
- Context
- 2,000,000 tok
Gemini 3.5 Pro is Google’s announced next flagship model, unveiled at Google I/O on 19 May 2026. Its headline features are a 2-million-token context window — the largest of any production frontier model, double the 1M on Gemini 3.5 Flash — and a built-in Deep Think reasoning mode for the hardest scientific, mathematical and coding problems (Tech Times).
It is still not generally available, nearly three months after announcement. As of 13 August 2026 Gemini 3.5 Pro remains in limited Vertex AI enterprise preview, having now missed three targets: its late-June date, a widely reported 17 July date, and a third in early August. On 21 July 2026 Google shipped three new models — Gemini 3.6 Flash, Gemini 3.5 Flash-Lite and Gemini 3.5 Flash Cyber — but pointedly not 3.5 Pro (TechCrunch). Bloomberg reported Google was working to improve the model’s capabilities particularly in coding, with testers citing hallucinations and inconsistent outputs; Alphabet’s shares fell 4.4% the day that report landed. Gemini 4 has since entered pretraining.
The delay is no longer just a scheduling story. In the first week of August 2026 Google restructured its AI leadership — Demis Hassabis stepped back from running Google DeepMind, co-founder Sergey Brin returned from retirement to drive AI commercialisation, and Jeff Dean left after 27 years — after a year of senior departures that Fortune ties directly to the stalled model. See the Google provider page for the full picture.
Official benchmarks and pricing have never been published, so almost everything known about 3.5 Pro is announced specification or expectation rather than confirmed fact; this page flags which is which. Until it ships broadly, Gemini 3.1 Pro remains Google’s GA flagship, with the fast Gemini 3.6 Flash line the workhorse beneath it — and with Gemini 4 in training, it is a fair question whether 3.5 Pro arrives in a form that still matters.
Quick specs
| Provider | |
| Announced | 19 May 2026 (Google I/O) |
| Status | Still not GA as of 13 August 2026 — limited preview after three missed targets |
| Context window | 2,000,000 tokens (announced) |
| Reasoning | Built-in Deep Think mode |
| Pricing | Not published (reports expect ~$15/$60 — unconfirmed) |
| Benchmarks | Never published |
| Successor status | Gemini 4 in pretraining |
| Best for | Very long-context analysis, hardest reasoning, deep multimodal work |
| Limitations | Not generally available; specs largely announced/expected, not verified |
What is known (and not) about Gemini 3.5 Pro
Google positions Gemini 3.5 Pro to absorb the use cases previously routed to its Ultra tier: the hardest reasoning tasks, deep multimodal work and very long-context analysis (Enterprise DNA). Two things are officially announced:
- A 2M-token context window — double Gemini 3.5 Flash’s 1M and the largest in any production frontier model.
- A built-in Deep Think reasoning mode — extended chain-of-thought for hard science, maths and coding, building on the Gemini 3 Deep Think approach.
Beyond that, the picture is provisional. Official benchmarks and pricing have not been released. One report cited internal data suggesting 10–15-point SWE-bench Verified gains over the 3.1 generation, but that is unverified (Tech Times); and reports expect pricing around ten times Gemini 3.5 Flash — roughly $15 input / $60 output per million tokens — but Google has not confirmed it (byteiota). Treat both as expectations, not facts.
Availability and rollout
Gemini 3.5 Pro was announced at Google I/O on 19 May 2026, when Sundar Pichai asked for “until next month”. Its launch has since slipped three times.
| Target | Outcome |
|---|---|
| Late June 2026 (Pichai’s “next month”) | Missed |
| 17 July 2026 (widely reported) | Missed — Google shipped three Flash models on 21 July instead, without 3.5 Pro |
| Early August 2026 | Missed — still in limited preview |
As of 13 August 2026 it remains in limited Vertex AI enterprise preview. Bloomberg reported the hold-up as a performance-goals problem centred on coding, and Fortune reported engineers attributing it to a strategic failure to prioritise coding ability from summer 2025 — precisely when agentic coding became the industry’s competitive battleground. Reports indicate 3.5 Pro will reach Google’s $20 AI Pro and $99.99 AI Ultra consumer plans first when it does land. Because it is not available, our best AI models ranking marks it “Unreleased” and no longer recommends it as the long-context pick.
How Gemini 3.5 Pro is expected to compare
A clean, same-harness comparison is not possible because Google has not published benchmarks. Directionally, Gemini 3.5 Pro is expected to extend Google’s lead on context length (2M tokens) and multimodal breadth, and was positioned to challenge the frontier — but the bar has moved during the delay. Claude Opus 5, GPT-5.6 Sol and Grok 4.6 have all shipped since it was announced, and Google’s strongest available model now ranks behind Anthropic, OpenAI, xAI, Meta and several Chinese labs on independent intelligence benchmarks. For verified, available models, see Gemini 3.1 Pro, Claude Opus 5 and GPT-5.6 Sol.
Known limitations
Not generally available. As of 13 August 2026 it is limited-preview only, having missed three targets; most users cannot access it.
It may be overtaken before it ships. Gemini 4 is in pretraining, and Google’s engineering focus has shifted; a model announced in May that arrives after its successor has begun training will face a very different frontier.
Specs are largely announced or expected. The 2M context and Deep Think mode are official; benchmarks and pricing are not, and circulating performance figures are unverified.
Coding is the reported weak point. Bloomberg attributes the delay to performance goals, particularly coding — the capability most buyers now weight most heavily.
Expected high price. If the ~$15/$60 expectation holds, it would be among the more expensive frontier models — though again, unconfirmed.
FAQ
Is Gemini 3.5 Pro available yet?
No. It was announced at Google I/O on 19 May 2026 but, as of 13 August 2026, is still in limited Vertex AI enterprise preview — not generally available, nearly three months later. It has missed three targets: its late-June date, a widely reported 17 July date, and a third in early August. Google’s 21 July model drop shipped three new Flash models without it. Gemini 3.1 Pro remains Google’s current GA flagship.
When is the Gemini 3.5 Pro release date?
Google has not given a confirmed public release date. The model was announced on 19 May 2026 and has missed three targets (late June, 17 July, and early August); as of 13 August 2026 it is still in limited enterprise preview. Bloomberg reported delays over performance goals, particularly coding, and Google has since begun pretraining Gemini 4 and restructured its AI leadership — so there is no reliable public signal on timing. Treat any specific date circulating online as unconfirmed until Google announces one.
Has Gemini 3.5 Pro been cancelled?
No. Google has not cancelled it, and it remains in testing. But it has missed three release targets, its successor Gemini 4 has entered pretraining, and the leadership of the division building it changed in August 2026 — so while it is not cancelled, it is reasonable to question how significant it will be when it does arrive.
What is special about Gemini 3.5 Pro?
Two announced features stand out: a 2-million-token context window — the largest of any production frontier model — and a built-in Deep Think reasoning mode for the hardest problems.
How much will Gemini 3.5 Pro cost?
Google has not published pricing. Reports expect roughly ten times Gemini 3.5 Flash — about $15 input / $60 output per million tokens — but this is unconfirmed and should not be relied on.
How good is Gemini 3.5 Pro?
Unknown officially. Google has not released benchmarks; one unverified report suggested 10–15-point SWE-bench Verified gains over the 3.1 generation. We will add confirmed figures once Google publishes them.
Last verified 13 August 2026. Gemini 3.5 Pro is announced but not generally available, having missed three targets (late June, 17 July and early August); its 2M context window and Deep Think mode are official, while benchmarks and pricing have never been published and the figures cited here are reported expectations, not confirmed facts. Confirm against Google’s official pages once it ships.