Google is targeting July 17, 2026 for Gemini 3.5 Pro, according to multiple third-party reports — but Google itself hasn't confirmed the date, specs, or pricing circulating online. What's confirmed is narrower: the model exists, and it slipped from June after Google reportedly scrapped an earlier base model and restarted pretraining. Arriving weeks after Claude Sonnet 5, GPT-5.6, and Grok 4.5 all launched, the gap between leak and confirmation matters more than usual for anyone evaluating a frontier model right now.
What's actually confirmed versus circulating
Confirmed: Gemini 3.5 Pro runs internally at Google, the release slipped from June to July, and the delay involved rebuilding the base model. Circulating but unconfirmed: a 2-million-token context window, a "Deep Think" slow-reasoning mode, a jump in frontend code generation quality, and leaked API pricing around $12–15 per million input tokens and $36–45 per million output tokens. Every one of those specifics is exactly the kind of detail that ends up wrong in leak culture — treating any of it as fact before Google publishes risks locking procurement decisions to numbers that could be materially different at general availability.
How to actually plan around an unconfirmed launch
Build your evaluation harness now, model-agnostic. If you already run your own benchmark suite against new models, point it at Gemini 3.5 Pro the day it ships instead of losing a week building tooling after the fact.
Don't renegotiate contracts based on leaked pricing. If the leaked numbers are even close, this positions Gemini 3.5 Pro as premium, not an automatic replacement for what you're running today.
Watch the frontend generation claim if you're in web tooling. Claude Fable 5's recent WebDev Arena win is a live benchmark; if Gemini's frontend claims hold up, that comparison resolves publicly and fast.
The bigger pattern: every frontier lab shipped in six weeks
Claude Sonnet 5, GPT-5.6 (as three variants), Grok 4.5, GLM-5.2, and now Gemini 3.5 Pro have all landed or queued within roughly six weeks of each other. That's not coincidental R&D timing — no lab can afford to let a competitor hold an uncontested capability or pricing edge for more than a few weeks. If you standardized on a single model provider a year ago and haven't revisited it, this compressed cadence is itself the strategic signal: the shelf life of "best model for the price" is now weeks, not quarters.
Conclusion
Gemini 3.5 Pro is real, delayed, and probably close — but everything specific about it right now is a leak. Build your evaluation pipeline this week so you're ready to test the actual model the day Google confirms it, rather than planning around numbers that might not survive contact with the GA announcement.