Leaked launch plans circulating this week suggest that Google DeepMind’s Gemini 3.5 Pro is targeting a general availability date of July 17, 2026, arriving nearly six weeks later than originally expected and landing in what could be the most competitive single week in AI modeling history – days after GPT-5.6 and Grok 4.5.
The leaks, which Google hasn’t officially confirmed, come primarily from industry reports and posts on X from developers tracking early access builds. A widely circulated thread laid out the core specs.
A complete rebuild, no tuning
According to the leaked details, Google has scrapped its original plan to build Gemini 3.5 Pro on the existing Gemini 2.5 Pro base and instead implemented an entirely new pre-training process – a decision that was reportedly responsible for the delay from an originally planned June launch.
The rebuilt model is set to ship with a 2 million token context window, twice the size of anything currently available in the frontier model space, and an expanded “deep think” reasoning mode that sits behind Google’s $250-per-month AI Ultra subscription tier.
Leaked benchmarks suggest a big jump in front-end and design capabilities in particular – better UI and layout assessment, cleaner generated interfaces and stronger SVG generation – alongside improved one-shot encoding and multi-stage reasoning performance. API prices are rumored to reach around $1.25 per million input tokens and $10 per million output tokens, positioning them at the premium end of the current market.
Competing leaks, a single date
Not every detail in circulation agrees. Several leak threads reportedly contradict each other regarding secondary specifications, and Google has not made any official comment on any of the numbers. However, the target date of July 17th has been repeatedly shown in various leaks and reports, with July 24th mentioned as a replacement date in case Google needs more time.
The timing matters beyond Google’s own roadmap. If Gemini 3.5 Pro lands as leaked, it would arrive the same week that Meta pushes its own Muse Spark 1.1 model and new developer API, and just days after the releases of OpenAI’s GPT-5.6 and xAI’s Grok 4.5 – compressing a normally staggered release calendar into a single, unusually dense stretch of frontier model launches.
For developers and companies considering which model to build on, mid-July 2026 is likely to be a rare time when the best available models from several major labs are all on the table within the same two-week window.
Google has not set an official announcement date and has not publicly responded to the leaked specifications.
Quick links:
