Gemini 2.5 Deep Think

2025-08-01 · Publisher : Google · Domain : Artificial Intelligence

Google rolls out Gemini 2.5 Deep Think to Google AI Ultra subscribers: a parallel-thinking reasoning mode, a faster variant of the model that won a gold medal at the International Mathematical Olympiad.

What changes

Parallel thinking: the model generates several ideas at the same time, revises or combines them before answering.
Everyday-use version, faster than the model that reached gold-medal level at IMO 2025; it reaches bronze level on this benchmark (internal evaluations).
Google also provides the full version, the gold-medal one, to a small group of mathematicians and academics.
Automatic use of tools (code execution, Google Search) and longer answers.
Improved content safety, but a stronger tendency to refuse harmless requests.

### Specifications - Availability: Gemini app, Google AI Ultra subscribers, by enabling "Deep Think" with 2.5 Pro; daily request count limited. - API: access planned for trusted testers, with no general availability date. ### Pricing Included in the Google AI Ultra subscription (subscription price not included here). No API pricing at launch. ### Benchmarks Google reports state of the art without tools on LiveCodeBench V6 and Humanity's Last Exam; the figures appear in an image in the post. ### Key takeaways A long reasoning mode reserved for the most expensive tier, with no open API in 2025.

Gemini 2.5 Deep Think

What changes

Links