Gemini 3 Pro
Gemini 3 Pro scored 1501 Elo on LMArena, 37.5% on Humanity's Last Exam, 91.9% on GPQA Diamond and 76.2% on SWE-bench Verified, and shipped in Search on day one.
Multimodal model with 1M-token context and new thinking-level and media-resolution controls. $2 / $12 per million tokens for prompts up to 200k. First Gemini model in Search (AI Mode) on launch day, plus the Gemini app, CLI and Antigravity.
- Date
- Tuesday, 18 November 2025
- Lab
- Google DeepMind
- Kind
- model
- Access
- closed API
- Price
- $2 in / $12 out per 1M tokens (prompts <=200k), preview, 2025-11-18
Figures
| Measure | Value | Measured by |
|---|---|---|
| LMArena | 1501 Elo topped leaderboard at launch | LMArena |
| Humanity's Last Exam (no tools) | 37.5% company-reported | company |
| GPQA Diamond | 91.9% company-reported | company |
| SWE-bench Verified | 76.2% company-reported | company |
| Terminal-Bench 2.0 | 54.2% company-reported | company |
Wikipedia describes it as a sparse mixture-of-experts model; not confirmed in the sources opened here. The Gemini API changelog lists gemini-3-pro-preview as shut down 2026-03-09 and redirected to 3.1 Pro.
Sources
- deepmind.google/blog/a-new-era-of-intelligence-with-gemini-3/
- deepmind.google/blog/start-building-with-gemini-3/
- blog.google/technology/ai/google-ai-updates-november-2025/
This record was checked against its sources on 6 October 2026. How we check