Google DeepMind delays Gemini 3.5 Pro to July 17 for architectural rebuild

Google DeepMind is trading speed for a bigger swing: rather than incrementally updating the existing Gemini 2.5 Pro framework, it is rebuilding Gemini 3.5 Pro from the architecture up, pushing the release to July 17. The stated goals are meaningful jumps in mathematical reasoning, SVG scene generation and image quality—areas where rivals have made noise—plus a claimed 2-million-token context window and a 'Deep Think Reasoning Layer' aimed at multi-step problems.

Mechanically, a full rebuild versus a fine-tune implies changes to the model's core structure, not just training data or RLHF, which is why it needs more time. A 2M-token context would match or exceed the largest windows on the market, and a dedicated reasoning layer suggests Google is productizing test-time compute the way competitors have.

Competitively the delay is risky timing: it slips into a week already crowded with GPT-5.6, Grok 4.5 and Anthropic's Sonnet 5, so Gemini 3.5 Pro arrives to a market that will have fresh frontier baselines to be measured against. Reviewers still rate Gemini's Pro models ahead of Apple's Siri AI for research, and Google's AI Ultra plan ($99.99/month) undercuts ChatGPT Pro by a cent. Skeptics will note that a rebuild-driven delay raises expectations it must then meet. Watch whether July 17 holds and whether the reasoning and SVG claims survive independent testing.