What happened
Google launched Gemini 3, its latest AI model, which the company claims performs exceptionally well in business operations and other tasks. The model achieved top scores across multiple benchmark tests, outperforming competitors including OpenAI's GPT-5.1 and Anthropic's Claude Sonnet 4.5. Google positioned the release as a significant advancement, with the model available across its product suite including Search on day one.
Why it matters
- Gemini 3 scored first place on 19 of 20 benchmarks tested against leading competitors, establishing it as the highest-performing model currently available.
- The model demonstrated a substantial performance gap in factuality benchmarks, with approximately 40% better performance than competitors on SimpleQA Verified, which measures hallucination rates.
- The widespread rollout to Google Search and other products means millions of users now have access to the highest-performing AI model available.
Key facts
- 1. Gemini 3 Pro achieved 37.4 on the Humanity's Last Exam benchmark, the highest score on record for that test.
- 2. The model earned approximately $5,500 on the Vending-Bench 2 business operations benchmark, compared to approximately $3,800 for the second-place model.
- 3. Gemini 3 Pro scored 93.8% on the GPQA Diamond benchmark and 45.1% on ARC-AGI-2 with code execution.
