ττ-bench Leaderboard

pronounced β€œhyper-tau-bench”

RankDeveloperDateHarnessReasoning
β€”
Human performance
β€”β€”
82.2%
β–Ά
#1
Claude Opus 5
Aug 31, 2026πŸ› οΈ Claude Codemax
23.9%
β–Ά
#2
GPT-5.6-sol
Aug 31, 2026πŸ› οΈ Codexxhigh
22.0%
β–Ά
#3
GPT-5.6-terra
Sep 1, 2026πŸ› οΈ Codexxhigh
18.0%
β–Ά
#4
Kimi K3
Sep 2, 2026πŸ› οΈ OpenCodemax
17.9%
β–Ά
#5
Kimi K3
Aug 31, 2026πŸ› οΈ Kimi Codemax
16.1%
β–Ά
#6
Claude Sonnet 5
Sep 2, 2026πŸ› οΈ Claude Codemax
14.9%
β–Ά

Submit Your Results

Have new results to share? Submit your developer configuration β€” coding harness and builder model β€” with a pull request that adds a submission.json under hyper-submissions/. The README's submission section has the required format and process.