Gemini 2.5 Deep Think
Google's parallel multi-agent reasoning mode — an IMO gold-medal variant of Gemini 2.5 Pro that explores many solutions at once.
About
Deep Think is Google's answer to the hardest reasoning problems: instead of one long chain of thought, it spawns multiple agents that attack a problem from different angles in parallel, then reconciles their work into a single answer. It plugs into tools like code execution and Google Search mid-reasoning, and Google says it produces substantially longer, more considered responses than standard Gemini.
The research variant hit the gold-medal standard at IMO 2025, and the shipping version in the Gemini app still reaches bronze-level on the 2025 IMO benchmark per Google's internal evals. It also posted 34.8% on Humanity's Last Exam without tools and 87.6% on LiveCodeBench v6. The catch is access and cost: it sits behind the $250/month Google AI Ultra tier, and the extra thinking time makes it slow by design.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Google-reported (blog.google, Aug 2025): 34.8% on Humanity's Last Exam (no tools), vs Grok 4 at 25.4% and o3 at 20.3%
- Same report: 87.6% on LiveCodeBench v6 coding benchmark
- Full research variant: gold-medal standard at IMO 2025; Gemini app release reaches bronze-level on the 2025 IMO benchmark (internal eval)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (TIER verified, metrics-v1). Evidence: Google DeepMind official launch (blog.google, Aug 1 2025) with HLE 34.8% and LiveCodeBench v6 87.6% figures; IMO 2025 gold-medal claim for the research variant; widely reported by TechCrunch and others. Strengths: flagship lab, detailed public benchmark disclosure, genuinely novel parallel-agent reasoning design. Docked for: (1) closed weights and gated behind a $250/mo tier; (2) headline numbers are vendor-reported on internal evals. License: proprietary.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 2.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Google DeepMind; launched 2025-08-01
- · IMO 2025 gold (research variant); HLE 34.8%, LiveCodeBench v6 87.6%
- · Benchmarks are vendor-reported