Gemini 3.5 Pro Release Date: Still No Launch as Gemini 3.7 Flash Ships

A week ago this page was a status check on Gemini 3.5 Pro: the model had surfaced in an anonymous Arena preview that Google never confirmed, the public model list was empty of it, and the question was when — not whether — it would ship. On August 14, day 87 of "next month," the API models page finally changed — but in the direction no one hoping for a flagship wanted. The new entry is Gemini 3.7 Flash, a stable model aimed at complex coding and agentic workflows, at an introductory $0.75/$3.75 per million tokens. Gemini 3.5 Pro still does not appear anywhere on that page. Three Flash generations have shipped since I/O — Gemini 3.5 Flash on May 19, Gemini 3.6 Flash on July 21, Gemini 3.7 Flash on August 13 — and the Pro model Google promised "next month" has missed every window since. On X the reading is hardening accordingly: as commentator @kimmonismus put it, "We got Gemini 3.5, 3.6 and now 3.7. At this point it's pretty clear that we won't see any 3.5 pro release." That is an inference from the release record, not a confirmation from Google — but it is the first time the record has visibly supported it.

The headline numbers first, because the verifiable picture changed for the first time in two weeks. Independently verifiable: Gemini 3.5 Pro still has no model ID in the Gemini API, no Vertex AI listing, no pricing, and no confirmed date — the newest Pro-class ID on Google's own models page is still gemini-3.1-pro-preview, dated February 19. Also independently verifiable as of August 14: gemini-3.7-flash is a live, stable model in Google's API with a 1-million-token input context. Disputed as ever: whether Gemini 3.5 Pro's absence means "still in testing" or "shelved." SemiAnalysis says the model was silently cancelled; Google calls the report "superficial" and says it remains in partner testing. What is new is that the Flash line's momentum — a third generation in twelve weeks — gives the second reading more support than it had a week ago, without making it true.

What the Arena preview actually led to

The signal this page was built around resolved exactly the way the previous version predicted it would. An anonymous Arena listing is a pre-release measurement step, not a release: the July 31 entry was live for roughly thirty minutes before being pulled, the August 5 backend-tag claim was a string in serving config, and the August 6 sighting was a running model served to anonymous testers. Google never confirmed the model's identity, and no release followed any of it. What shipped instead, twice over, was more Flash. The Arena preview led to nothing; the Flash line moved on to a second and then a third generation while the flagship stayed off every list.

The SemiAnalysis report — and Google's pushback

The new information in the last update was a report, not a release, and nothing has changed that. On August 11, SemiAnalysis — a semiconductor and AI research firm whose analyses are widely circulated in the model-industry press — published a report arguing that Google had silently cancelled Gemini 3.5 Pro rather than keep waiting for it. Its claims, all attributed to SemiAnalysis and none independently confirmed: the release-candidate checkpoint Google had been preparing (referred to in the report as "gemdelta") disappeared from internal systems days before a scheduled release; the model's coding ability was still falling short of internal targets; and SemiAnalysis estimated the model's overall capability at roughly the level of Anthropic's Claude Opus 4.5, which shipped in November 2025 — putting Google's flagship about half a year behind the frontier. The report also argued that Gemini as a family had slipped to eighth or ninth place on the Artificial Analysis Intelligence Index, and that Google's attention was already shifting to Gemini 4.

Google disputes the report. Its official position, unchanged since July 21, is that Gemini 3.5 Pro is "currently testing with partners" and will be made broadly available "as soon as it's ready." In response to the SemiAnalysis report, Logan Kilpatrick, Google's product lead for the Gemini API, called the analysis "superficial" and pointed to the rapid rollout of Gemini 3.6 Flash as evidence of execution. The August 13 launch of Gemini 3.7 Flash is now Google's best execution evidence — it is real, it is stable, and it is not Gemini 3.5 Pro. Google DeepMind's Gemini Pro page still carries a "3.5 Pro coming soon" label. None of that confirms the model will ship — that eyebrow has now been up for nearly twelve weeks — but it is the difference between "reported cancelled" and "cancelled," and the two should not be confused.

The leadership reshuffle that changed who decides

Two days before the SemiAnalysis report, Google announced a sweeping reshuffle of DeepMind's leadership. Demis Hassabis, the lab's co-founder and CEO, stepped back from day-to-day control to become chairman of Google DeepMind and Alphabet's first chief scientist. Koray Kavukcuoglu, previously DeepMind's CTO, was promoted to senior vice president and took over day-to-day control of the lab — including Gemini model development — reporting directly to Sundar Pichai. Co-founder Sergey Brin, who has held no formal title since 2019, is widely reported to be re-engaging in Gemini strategy. Alphabet's stock fell about four percent on the announcement. Separately, Jeff Dean, Sanjay Ghemawat, Oriol Vinyals and Quoc Le — four senior figures across the Gemini effort — left to found a new lab called Discovery Loop.

Why this matters for a release-date question: the people who owned the "when does it ship" decision changed mid-stream, and Google's public posture has pivoted. On the earnings call in late July, Pichai defended the delay, acknowledged coding as an area needing improvement, and framed the roadmap around Gemini 4 and a near-monthly cadence of releases. That near-monthly cadence is now visible in the Flash line — 3.5 Flash in May, 3.6 Flash in July, 3.7 Flash in August — while the model this page tracks is discussed, when it is discussed at all, as a stepping stone rather than a destination.

Made by Google came and went

The calendar's last rumored date for a Gemini 3.5 Pro announcement was August 12, the day of Google's Made by Google hardware event. It came and went. The keynote unveiled the Pixel 11 lineup, the Pixel Watch 5, a new Pixel Tag, a set of "Gemini Intelligence" features and thirteen new connected apps, and it opened with the fact that Gemini now passes one billion monthly users. There was no Gemini 3.5 Pro — not even as a teaser. And the next day, August 13, Google shipped Gemini 3.7 Flash instead, which tells you which line the company is actually feeding.

One wrinkle to keep in mind before reading too much into that silence: the consumer app can get a model before the API does. A Pro announcement aimed at consumers could still happen without a model ID in the API. But twelve weeks into a model that Google promised "next month," an unmentioned keynote followed by a Flash launch is not evidence of progress — it is another window closing, and the slot being filled by someone else.

What we could verify on August 14

The same four checks this page has run for two weeks, re-run today, produce one changed result. First, Google's Gemini API models page: for the first time since the 3.5 family landed, there is a new entry — Gemini 3.7 Flash, marked "New Stable" with the description "our latest and most capable Flash model," linking to the stable model ID gemini-3.7-flash. The 3.5-generation entries remain Gemini 3.5 Flash, Gemini 3.5 Flash-Lite and Gemini 3.5 Live Translate, and there is still no Gemini 3.5 Pro and no preview-suffixed variant of it. Second, the Vertex AI release notes: no Gemini 3.5 Pro mention. Third, Google DeepMind's Gemini Pro page: the "coming soon" eyebrow is still up. Fourth, routers and aggregators: no gemini-3.5-pro listing anywhere, ours included — OrcaRouter does not carry a model that has no endpoint.

The conclusion is the same, and now sharper: the model is not callable. If gemini-3.5-pro is not in Google's model list, a hard-coded string means a failed request in production. What changed is that the page is no longer static. Google updated it on August 13 to add Gemini 3.7 Flash and chose not to add the model this page tracks. That is a decision visible in the public record, and it is the strongest sign yet that "waiting on Pro" is not a strategy.

The timeline, updated to day 87

It is worth re-running the full timeline, because the record of missed windows is the most reliable predictor available. May 19: Google announces the Gemini 3.5 family at I/O and ships Gemini 3.5 Flash; on Pro, Pichai says it is "already being used internally" and Google looks forward to rolling it out "next month." June: nothing. July 16: Bloomberg, citing ten current and former employees, reports the model is months behind schedule with coding short of internal goals. July 17: a widely-circulated rumored date passes. July 21: Google ships Gemini 3.6 Flash, Gemini 3.5 Flash-Lite and Gemini 3.5 Flash Cyber; Kilpatrick says Pro is "currently testing with partners" and Google "hopes to land soon." July 31: a gemini-3.5-pro entry appears on Arena and is pulled after roughly thirty minutes. August 5: a backend tag is reported, and Google reshuffles DeepMind's leadership. August 6: the Arena sighting this page covered. August 11: the SemiAnalysis cancellation report. August 12: Made by Google passes without a mention. August 13: Google ships Gemini 3.7 Flash — stable ID gemini-3.7-flash, 1M-token context, intro pricing $0.75/$3.75 per million tokens through the end of 2026 — the third Flash-generation release since I/O, and the first change to the API models page in two weeks.

That is day 87 after "next month." On the prediction markets, the July 31 market on a Gemini Pro release resolved no, and money has since been shifting toward no release by the end of August, with no single date drawing meaningful weight. The 3.7 Flash launch is consistent with that drift: the market is pricing a Pro release as increasingly unlikely, and the release record keeps cooperating.

What this means for developers

If Gemini 3.1 Pro is failing your hardest tasks, you are ceiling-blocked, and waiting on a model whose status is now disputed is the weakest plan on the table — several frontier models shipped during the twelve weeks since I/O, and the honest comparison has moved on. If Gemini 3.1 Pro covers your work, the wait was previously nearly free; it still is, but it now carries a new risk, which is that the payoff never arrives. Test Gemini 3.7 Flash as a cheaper, current cover either way.

Google's own numbers make that comparison sharper than the version numbers suggest. Current pricing on Google's list: Gemini 3.1 Pro is $2.00 per million input tokens and $12.00 per million output at a 1-million-token context; Gemini 3.6 Flash is $1.50/$7.50; Gemini 3.5 Flash-Lite is $0.30/$2.50. New this week: Gemini 3.7 Flash at $0.75/$3.75 — half of Gemini 3.6 Flash's list price — on an introductory rate reported to run through the end of 2026 and then double back to $1.50/$7.50 on January 1, 2027. And at I/O, Google positioned Gemini 3.5 Flash as "frontier-level intelligence" that beat the then-current Pro on coding and agentic benchmarks — a vendor claim, unreproduced, but it is the vendor's own stated rationale for buying the cheaper tier. On its own figures, for many coding and agent workloads the current Flash generation is now both cheaper and newer than the current Pro generation; the wait for Pro only makes sense if you are genuinely ceiling-blocked. Gemini 3.1 Pro's own headline benchmarks — 80.6% SWE-Bench Verified, 94.3% GPQA Diamond, 77.1% ARC-AGI-2, 84.9% MRCR v2 at 128K, Elo 2887 on LiveCodeBench Pro — are vendor-reported, unreproduced by independent labs.

Keep shipping on Gemini 3.1 Pro if you need Pro-tier work today. Across seven days of OrcaRouter traffic the model has shown a p50 time-to-first-token of 7.43 seconds and a p95 of 10.00 seconds over 2.1 million tokens of production requests — a current datapoint if you are sizing it. Do not add a gemini-3.5-pro string anywhere in code; keep model strings in configuration; run an A/B rather than a migration; and do not chase partner access, because partner builds change most before GA. On OrcaRouter, Gemini 3.1 Pro and Gemini 3.6 Flash are both reachable through the same key, which makes the A/B a config edit rather than a procurement exercise, and automatic failover means a sudden status change on either one is a routing decision instead of an incident. The moment Google publishes a price for Gemini 3.5 Pro, the pass-through pricing on our side shows the provider's list number the same day, with no markup to wait out — but that day has not arrived, and "there is nothing to route to yet" remains true.

If it ever ships, where it will show up first

The ordering from the previous version is unchanged, and it is the best antidote to Arena-based speculation. In order: the Gemini API model list and release notes; the Vertex AI release notes; Google DeepMind's Pro page losing its "coming soon" eyebrow and gaining a model card; a post on blog.google with the headline numbers; and routers and aggregators, ourselves included, within hours of the API listing. An Arena appearance is deliberately not on that list, and nothing about this week changes that — the August 13 update to the models page shows exactly how Google announces a model when it actually ships one.

The one sequencing wrinkle worth repeating: the Gemini consumer app can get a model before the API does, which is how "it's out" and "there is no model ID" can both be true for a few days. If you see people using a model in the consumer app, check the model list before believing the API is live.

The honest read

A week ago the position was: the model exists and is being wired up, but the artifact-before-launch signals were worth roughly nothing as a date. That assessment aged well — the Arena preview led nowhere, exactly as predicted. What is new is that the release-date question now has two data points it lacked before. On the verifiable side, the API models page changed on August 13 for the first time in two weeks — and the addition was Gemini 3.7 Flash, not Gemini 3.5 Pro. On the interpretive side, the community reading has hardened: three Flash generations shipped in twelve weeks while the promised flagship missed every window, and a growing number of observers now read that pattern as "there will be no Gemini 3.5 Pro release." That is an inference from the release record, not a confirmation from Google — SemiAnalysis's cancellation claim remains disputed, and Google still says "partner testing" — but it is an inference the record now supports more strongly than the alternative.

What a reader can actually decide is narrower, and the last two weeks have only made it clearer. Do not hard-code the string. Do not freeze a roadmap on the model. If you need frontier Pro-tier work, Gemini 3.1 Pro is still the current, shipping choice and its price is stable; if you are ceiling-blocked, the field has moved on and the waiting has a real opportunity cost. And test the Flash tier you can call today: by Google's own list prices, Gemini 3.7 Flash is now the cheapest current generation at half the price of Gemini 3.6 Flash, which makes the "wait for Pro" argument the thinnest it has been since I/O. The release date of Gemini 3.5 Pro is unknown — and for the second week running, the honest reason is that it may not have one.

Questions people are actually asking

Has Gemini 3.5 Pro been released?

No. As of August 14 there is no gemini-3.5-pro model ID in the Gemini API or Vertex AI, no pricing, and no confirmed date. The newest Pro-class ID on Google's models page is gemini-3.1-pro-preview, dated February 19. What did change on August 13 is that the same page gained Gemini 3.7 Flash — the model that shipped in Pro's place.

Did Google cancel Gemini 3.5 Pro?

Not confirmed. SemiAnalysis reported on August 11 that the model was silently cancelled, citing coding shortfalls and a release-candidate checkpoint that disappeared. Google's product lead called the report "superficial," and Google's official position remains that the model is in partner testing. The August 13 launch of Gemini 3.7 Flash — a third Flash generation in twelve weeks — is consistent with the "superseded" reading, but it is not confirmation; the verifiable facts still fit both a long delay and a quiet shelving.

Was Gemini 3.5 Pro announced at Made by Google?

No. The August 12 keynote covered the Pixel 11 lineup, Pixel Watch 5, Gemini Intelligence features and new connected apps, with no mention of Pro — and the next day Google shipped Gemini 3.7 Flash instead. The consumer app could still receive a model before the API, so the keynote is not proof of cancellation; but the last rumored date on the calendar is now two days gone, and the announcement slot was filled by a different model.

Should I keep waiting for Gemini 3.5 Pro?

It depends on which wall you are hitting. If Gemini 3.1 Pro is failing your hardest tasks, the wait now has a serious opportunity cost, and several frontier models shipped during the twelve weeks since I/O. If Gemini 3.1 Pro covers your work, test Gemini 3.7 Flash as the current cover — at $0.75/$3.75 it is half the list price of Gemini 3.6 Flash, which makes the wait cheaper to live with — but treat any plan that depends on Gemini 3.5 Pro arriving as conditional.

Will Gemini 3.5 Pro cost more than Gemini 3.1 Pro?

Unknown. There is no leaked price and, unusually, not even a rumored one. Google's recent pattern has been efficiency framing rather than rate cuts, so flat-to-higher remains the safer assumption to budget against — but that is a budgeting assumption, not a figure.