Gemini 3.5 Pro Targets July 17 Launch with 2 Million Token Context Window and Deep Think Reasoning

My take: According to multiple reports published in the past few days, Google DeepMind is targeting July 17 for the Gemini 3.5 Pro launch. What stands out is not the date but the core decision: they scrapped the previous architecture and rebuilt from scratch with a new pretraining run. That explains the months of delays and also signals this is not an incremental update.

The most concrete data circulating: a 2 million token context window (double that of Gemini 2.5 Pro), a new reasoning layer called Deep Think for multi-step problems, and improvements targeting math, SVG generation, and frontend coding. Leaked pricing estimates put input costs at around $12 to $15 per million tokens.

To be clear: the July 17 date and technical specifications are circulating as leaks, not as an official Google announcement. Until an official model card or API documentation appears, treat these numbers with appropriate skepticism. This model has a track record of slipped dates.

For anyone designing systems with language models: is it worth waiting for Gemini 3.5 Pro before committing to the architecture of a new project, or better to build with what is available today and migrate later if it makes sense?