Gemini 3.1 Pro Preview achieves 46.4% on Humanity's Last Exam, the highest among 10+ top models?

Gemini 3.1 Pro Preview achieves 46.4% on Humanity's Last Exam, the highest among 10+ top models.

Claude Fable 5 leads SimpleBench with 81.9%, outperforming GPT-5.5 Pro by 5 percentage points?

Claude Fable 5 leads SimpleBench with 81.9%, outperforming GPT-5.5 Pro by 5 percentage points.

Independent benchmarks (Epoch, Scale) provide unbiased scores, unlike vendor self-reports.

Models & Releases

Lmcouncil June 24, 2026

⚡Gemini 3.1 Pro Preview leads hardest exam; Claude Fable 5 tops common sense.

Deep Dive

Key Points

Gemini 3.1 Pro Preview achieves 46.4% on Humanity's Last Exam, the highest among 10+ top models.
Claude Fable 5 leads SimpleBench with 81.9%, outperforming GPT-5.5 Pro by 5 percentage points.
Independent benchmarks (Epoch, Scale) provide unbiased scores, unlike vendor self-reports.

Independent benchmarks reveal which models truly handle complex reasoning and common sense, guiding enterprise AI selection and investment.