Gemini 3.5 Pro leak reveals 'Extra High' reasoning tier
A leaked screenshot indicates Google's upcoming Gemini 3.5 Pro will feature an "Extra High" thinking tier. This suggests Google is introducing user-controlled variable inference compute, allowing the model to spend extended time on multi-step logic.
This leak points to the next battleground in the foundation model wars: user-controlled inference scaling.
- –The "Extra High" mode likely mirrors long-form reasoning, trading instant speed for deliberate execution on complex logic and architecture tasks
- –The developer community is split on whether this is a pure capability boost or a mechanism to throttle default requests and manage compute costs
- –With Gemini 3.5 Pro expected in June, this variable-thinking approach could make it a highly cost-efficient backend for autonomous agent workflows
DISCOVERED
2h ago
2026-05-26
PUBLISHED
5h ago
2026-05-26
RELEVANCE
AUTHOR
Independent-Wind4462