Leaderboards

Models ranked per benchmark, and the composite modhub Index — a plain average across every benchmark we track. Estimates are marked with ~.

modhub Index — overall standings

1OpenAI o4-mini80.92Qwen3-Max80.53GPT-5.577.24Gemini 3 Pro73.65Gemini 2.5 Pro70.96GPT-570.77Qwen3-235B-A22B69.08OpenAI o368.99MiniMax M268.910DeepSeek V3.268.411DeepSeek R1 (0528)67.812Kimi K2 Thinking67.613Claude Sonnet 4.565.514GLM-4.663.915GPT-5.159.6

Models need results on at least 4 tracked benchmarks to qualify.

By benchmark

Coding

SWE-bench Verified

  1. 1Claude Fable 5~95%
  2. 2Claude Opus 4.888.6%
  3. 3GPT-5.582.6%

Reasoning

GPQA Diamond

  1. 1GPT-5.5~92%
  2. 2Gemini 3 Pro91.9%
  3. 3Claude Opus 4.8~91%

Math

AIME 2025

  1. 1GPT-5.5~99%
  2. 2Gemini 3 Pro95%
  3. 3GPT-594.6%

Reasoning

HLE

  1. 1Gemini 3 Pro37.5%
  2. 2GPT-5.5~35%
  3. 3GPT-5.1~26.5%

Agentic

Terminal-Bench

  1. 1Claude Opus 4.8~63%
  2. 2Claude Opus 4.559.3%
  3. 3Gemini 3 Pro54.2%

Multimodal

MMMU

  1. 1Gemini 3 Pro~87%
  2. 2GPT-584.2%
  3. 3OpenAI o382.9%

Knowledge

MMLU-Pro

  1. 1GPT-5~87%
  2. 2Gemini 2.5 Pro86.2%
  3. 3Qwen3-Max~85.2%