Gemini 3.5 Pro pushed to July 17
Google scrapped the 2.5 Pro architecture and rebuilt Gemini 3.5 Pro from scratch, adding a 2M-token context and Deep Think mode for a July 17 release.
Google DeepMind pushed Gemini 3.5 Pro to July 17 after scrapping the 2.5 Pro base architecture and restarting pre-training from scratch. The decision was not a tune-up: the company ran a full rebuild targeting three specific performance gaps against GPT-5.6 and Fable 5 in mathematical reasoning, SVG scene generation, and image quality.
The rebuilt model ships with a 2-million-token context window and a Deep Think mode, an extended inference-time compute layer that spends additional reasoning cycles before responding. Latency and token usage go up in exchange for accuracy on hard problems. As of July 9, the model sits in limited Vertex AI preview with enterprise allowlist access; general availability is expected at launch on the 17th.
The delay follows a rough June for Google. Four senior DeepMind researchers departed in a single week, including Gemini co-lead Noam Shazeer and Nobel laureate John Jumper, both of whom moved to competitors. Alphabet shares fell roughly 5% on June 22, erasing around $225 billion in market value.
What it means for you
The 2M-token context is the largest of any production frontier model right now, which matters for long-document pipelines and multi-file analysis tasks that smaller windows hit a wall on. If you are already building on Gemini 3.5 Flash’s computer-use tooling, that release is unaffected by this delay. Pricing and official benchmarks have not been disclosed; wait for July 17 before making routing calls based on math or visual-generation performance.