Gemini 3.5 Pro remains unreleased as Google’s launch promises slip

As of August 25, 2026, Gemini 3.5 Pro is not publicly available. Google’s official model documentation lists Gemini 3.7 Flash as its latest and most capable Flash model, while DeepMind’s model page still describes 3.5 Pro as “coming soon.” There is no public Gemini 3.5 Pro model identifier, API listing, pricing page, or confirmed release date in the available Google documentation.

That makes the clearest status update simple: Google has acknowledged the model, but it has not shipped it—and every specific launch date circulating outside Google should be treated as unconfirmed.

What Google actually promised

At Google I/O on May 19, Google said Gemini 3.5 Pro was already being used internally and would be rolled out “next month.” That created an expected June 2026 launch window.

Google did release Gemini 3.5 Flash at I/O. The Pro model, however, did not follow in June. A July 17 date then circulated through third-party reporting, but contemporary checks found no official Google announcement, model card, pricing page, or gemini-3.5-pro entry supporting it as a firm commitment. 511

The July 17 target also passed without a public launch. Reports subsequently described further slippage beyond July 21 and into early August, but those dates were likewise not formal Google commitments. 101516

Why the model is reportedly delayed

The explanation in circulation is more serious than an ordinary launch-date change. Reports attributed to current or former employees say the model struggled to meet internal quality standards, particularly in coding, hallucination rates, and real-world reliability. One account also described a late-June training-data change that made results worse rather than better. 61013

Other reports claim Google scrapped an earlier base model and moved to a deeper pre-training cycle, with some accounts characterizing the work as a near-complete retraining. Reports have also mentioned senior researcher departures and a possible structural problem. 12

These details should not be presented as settled facts. They come from third-party publications and unnamed-source reporting, not from a Google model card, engineering post, or launch announcement. Even the stronger “started over” language is disputed by reporting that confirms only a delay and additional testing. 514

What prediction markets indicate

Prediction markets are useful as sentiment indicators, not as official release guidance. Their prices change as traders react to reports, and different contracts may define a qualifying release differently.

The market view became markedly less optimistic during August:

• A late-July Polymarket snapshot implied about a 6.7% chance of a qualifying Pro release by August 31. 18
• An August 17 snapshot priced “no release by September 30” at roughly 73%. 17
• Earlier August market data put the chance of a release by September 30 at about 36%, while another snapshot placed the chance of a release by October 31 at 70.5%. 21
• A separate August 19 report described only a 10% chance of release by the end of August, with the odds not reaching 50% until September 30 and rising to 71% by October 31. 3

The changing numbers matter more than any single percentage. They show traders moving from a near-term launch assumption toward an autumn release window. The available evidence does not independently substantiate the precise claim that more than $1.1 million had been wagered, so that figure should not be treated as verified.

Google’s public fallback is Gemini 3.7 Flash

While 3.5 Pro remains absent, Google released Gemini 3.7 Flash with a launch date of August 13, 2026. Google describes it as a generally available, high-efficiency model for complex coding, agentic workflows, and reliable multi-step execution. Its documented model ID is gemini-3.7-flash.

Google’s own positioning is important: 3.7 Flash is a high-efficiency “workhorse” that bridges the gap between deep-reasoning Pro models and faster, lower-cost Flash-Lite models. That makes it a substantial production model, but not proof that it replaces the missing 3.5 Pro tier.

For developers, the practical distinction is therefore clear. Gemini 3.7 Flash is available now with published documentation and pricing, while Gemini 3.5 Pro remains a future product with no public contract around its capabilities, cost, or availability. Google lists an introductory price of $0.75 per million input tokens and $3.75 per million output tokens for 3.7 Flash.

What the delay means for Google

The delay creates a product-positioning gap. Google can offer a capable Flash model for coding and agentic work, but customers specifically waiting for a premium Pro model still have no public Gemini 3.5 Pro specification or launch commitment to evaluate.

It also makes comparisons with competing AI systems difficult to substantiate. Claims that Gemini 3.7 Flash definitively trails a particular OpenAI model on coding or complex reasoning require a defined model version, benchmark, prompting method, tool setup, cost assumptions, and independent reproduction. The available sources do not provide enough authoritative benchmark evidence for a categorical winner.

The most defensible conclusion is narrower: Google has shipped a strong, efficient Flash workhorse while its promised Pro model remains delayed and under-specified. Until Google publishes a model card, API identifier, pricing, and availability announcement, the reported rebuild narrative and all future launch dates remain provisional.