Eight days remain before June 30, and Gemini 3.5 Pro is still in limited Vertex AI enterprise preview. Sundar Pichai stood on the I/O stage on May 19 and told the room the model was already in internal use and would roll out "next month," a phrasing that, per TechCrunch's account, drew audible groans. A month later, the public Gemini app and AI Studio have nothing. Polymarket has the pre–June 30 ship at roughly 50–55%, which is what a coin flip looks like when traders are pricing in Google's reputation for slipping its own launch windows.

The release notes tell the cleaner story. As of this week, Gemini 3.5 Flash is generally available across Global, US, and EU multi-regions, enabled by default on June 8, with the feature toggle removed on June 16. Gemini 3.1 Pro is still listed under "Limited Availability." Gemini 3.5 Pro has no GA date at all. Flash, the smaller sibling, shipped on schedule. The flagship didn't.

What's on offer, then, isn't a model. It's a spec sheet, mostly from third parties.

Codersera's launch guide describes a 2-million-token input context window (Flash carries 1M), an extended-reasoning mode called Deep Think, frontier multimodal capabilities, and a positioning shift in which Pro absorbs workloads previously routed to an Ultra tier. Grow Wing Assistant reports Deep Think will be gated to the $250/month Ultra plan; the standard $20/month Pro subscribers get the 2M context and the base model but not the o3-style reasoning track. Techtimes notes, with appropriate hedging, that "June" is the target, not a guarantee.

The benchmark numbers circulating are entirely for models that already exist. On MRCR v2 long-context retrieval at 128k tokens, Gemini 3.1 Pro scores 84.9% and 3.5 Flash scores 77.3%; at 1M tokens both collapse to roughly 26%. On SWE-Bench Pro, Claude Opus 4.7 leads at 64.3%, GPT-5.5 at 57.7%, with 3.5 Flash at 55.1% and 3.1 Pro at 54.2%. Flash also posts 76.2% on Terminal-Bench 2.1 and 83.6% on MCP Atlas. None of these are 3.5 Pro numbers. They're the surrounding furniture.

Every previous Gemini model has launched through a single benchmark-grid post on blog.google. Until that post appears, the 2M-context-plus-Deep-Think narrative is Google's framing of a product it hasn't released, propagated by spec-roundup sites quoting one another. The competitive read is sharper than the calendar one: OpenAI and Anthropic both have shippable frontier models in market. Google has a keynote slide and a Vertex preview. That gap, not the eight days, is the story.

Sources

https://techcrunch.com/2026/05/19/with-gemini-3-5-flash-google-bets-its-next-ai-wave-on-agents-not-chatbots/
https://docs.cloud.google.com/gemini/enterprise/docs/release-notes
https://growwingassistant.com/ai-news/gemini-3-5-pro-release-date-june-2026-every-confirmed-spec-pricing-when-it-drops/
https://www.techtimes.com/articles/317919/20260606/google-gemini-35-pro-nears-june-launch-2-million-token-context-deep-think-reasoning.htm
https://codersera.com/blog/gemini-3-5-pro-launch-guide-2026/