Leaderboard history

Leaderboard history

Leaderboards move. This is the archive: who leads each core benchmark right now, and which earlier scores have been replaced.

Current leaders by benchmark

GPQA Diamondtop 3 now
  1. 1. GPT-6 Astra96%
  2. 2. Gemini 3.1 Pro94.3%
  3. 3. GPT 5.6-Sol94.1%
HLEtop 3 now
  1. 1. Claude Opus 5.564.4%
  2. 2. Claude Opus 556.3%
  3. 3. GPT-6 Astra54.7%
Terminal-Bench 2.1top 3 now
  1. 1. DeepSeek V4.1-Flash 091090.6%
  2. 2. GPT-6 Astra89.9%
  3. 3. Claude Opus 589.1%
DeepSWE v1.1top 3 now
  1. 1. Gemini 4 Argon77.9%
  2. 2. DeepSeek V4.1-Flash 101777%
  3. 3. DeepSeek V4.1-Flash 091074.2%
NL2Repo-Benchtop 3 now
  1. 1. Claude Opus 575.3%
  2. 2. DeepSeek V4.1-Flash 101768.2%
  3. 3. DeepSeek V4.1-Flash 091065.4%
Automation-Benchtop 3 now
  1. 1. DeepSeek V4.1-Flash 101760.5%
  2. 2. DeepSeek V4.1-Flash 091054.8%
  3. 3. Gemini 4 Argon51.3%

Overwritten scores

Nothing has been overwritten yet.

Scores are collected from named primary sources; see each model page for the exact citation.