What maximum INPUT context window will Gemini 3.5 Pro officially advertise at public release?

rationale:

As of mid-June 2026, Gemini 3.5 Pro has been teased but not officially released, making its final public context window a tight race between the established 1M status quo (50%) and a rumored 2M upgrade (46%).

The slight edge goes to the 1M bucket (1,048,576 tokens) due to strict family-wide consistency and historical precedent for under-delivery. Currently, Gemini 3.1 Pro 2 sources and the same-generation Gemini 3.5 Flash 2 sources both officially advertise exactly 1,048,576 input tokens. Crucially, Google has previously missed expectations on context size: Gemini 2.5 Pro was initially advertised with '2M coming soon' but stabilized at 1M for General Availability blog.google. Furthermore, Google's explicit strategy for peak capability in the 3.x generation centers on scaling inference-time compute (Deep Think/effort levels) rather than raw context expansion.

However, a 2M context window remains highly credible at 46%. Pervasive secondary reports and leaks specifically target 2M for 3.5 Pro 3 sources. This aligns with past Pro-tier capabilities—both Gemini 1.5 Pro developers.googleblog.com and 2.0 Pro Experimental blog.google offered 2M. Upgrading to 2M provides an obvious vector for product differentiation against the 1M 3.5 Flash. The 46% assessment reflects the tension between this logical market positioning and the fact that current official sources deliberately omit a confirmed context figure 2 sources.

Options above 3M are highly unlikely (4%). While Google has tested prior architectures up to 10M tokens blog.google, there is no credible leak or strategic indicator suggesting a jump of this magnitude for the official 3.5 Pro public launch.