Leaderboards

Benchmarks

Every score carries its origin.

Reasoning374Coding337Math221Agentic375Chat636Speech778
Arena Agent55Arena Agent · Recovery55Arena Agent · Steerability55Arena Agent · Task outcome55Arena Agent · Tool use55LiveBench Agentic Coding58SWE-bench Verified42

Arena Agent · Recovery

Inverse-propensity score for getting back on track after a command fails — higher is better
Provenance55 modelsscores Aug 11, 2026 – Sep 25, 2026
1Claude Opus 5AnthropicScore0.126ScoredSep 25, 20262Claude Fable 5.1AnthropicScore0.117ScoredSep 25, 20263Claude Opus 4.8AnthropicScore0.1ScoredSep 5, 20264Claude Opus 4.6AnthropicScore0.086ScoredSep 5, 20265Claude Fable 5AnthropicScore0.085ScoredSep 25, 20266GPT-6 SolOpenAIScore0.084ScoredSep 25, 20267Claude Opus 4.7AnthropicScore0.076ScoredSep 5, 20268DeepSeek V4.1 FlashDeepSeek763BScore0.074ScoredSep 25, 20269Claude Sonnet 5AnthropicScore0.066ScoredSep 25, 202610Hy4 previewTencent780BScore0.065ScoredSep 25, 202611GPT-6 AstraOpenAIScore0.065ScoredSep 25, 202612Muse Spark 1.2MetaScore0.063ScoredSep 25, 202613Grok 4.5xAIScore0.061ScoredSep 25, 202614Grok 4.6xAIScore0.058ScoredSep 25, 202615Inkling SmallThinking Machines266BScore0.057ScoredSep 25, 202616Grok 4.7xAIScore0.057ScoredSep 25, 202617GPT-5.5OpenAIScore0.055ScoredSep 25, 202618Muse Spark 1.3MetaScore0.054ScoredSep 25, 202619Claude Sonnet 4.6AnthropicScore0.05ScoredSep 15, 202620Kimi K3Moonshot AI2.8TScore0.043ScoredSep 25, 202621GPT-5.4OpenAIScore0.03ScoredSep 25, 202622Qwen3.8 MaxQwenScore0.028ScoredSep 25, 202623GPT-5.6 SolOpenAIScore0.028ScoredSep 25, 202624GPT-5.6 TerraOpenAIScore0.024ScoredSep 25, 202625DeepSeek V4 ProDeepSeek1.6TScore0.024ScoredSep 25, 202626GPT-5.6 LunaOpenAIScore0.022ScoredSep 25, 202627DeepSeek V4 FlashDeepSeek291BScore0.015ScoredSep 15, 202628GLM 5.2Z.ai753BScore0.011ScoredSep 25, 202629InklingThinking Machines952BScore0.01ScoredSep 25, 202630MiniMax M3MiniMax427BScore0.009ScoredSep 25, 202631Gemini 3.8 FlashGoogleScore0.003ScoredSep 25, 202632GLM 5.3Z.aiScore−0.008ScoredSep 25, 202633Muse Spark 1.1MetaScore−0.018ScoredSep 25, 202634Qwen3.7 MaxQwenScore−0.018ScoredSep 25, 202635Qwen3.8 FlashQwenScore−0.026ScoredSep 25, 202636Gemini 3.7 FlashGoogleScore−0.03ScoredSep 25, 202637Gemini 3.6 FlashGoogleScore−0.036ScoredAug 11, 202638Kimi K2.7 CodeMoonshot AI1.1TScore−0.039ScoredSep 5, 202639Qwen3.7 PlusQwenScore−0.054ScoredSep 25, 202640Hy3Tencent299BScore−0.057ScoredSep 25, 202641GLM 5.3 FlashZ.ai321BScore−0.057ScoredSep 25, 202642MiMo-V2.5-ProXiaomi1TScore−0.058ScoredSep 25, 202643Gemini 3.5 FlashGoogleScore−0.062ScoredSep 5, 202644Qwen3.8 27BQwen27.8BScore−0.062ScoredSep 25, 202645GLM 5.1Z.ai754BScore−0.062ScoredSep 5, 202646Kimi K2.6Moonshot AI1.1TScore−0.096ScoredSep 5, 202647Mistral Medium 3.5Mistral AI128BScore−0.179ScoredSep 25, 202648Gemini 3.1 Pro Preview Custom ToolsGoogleScore−0.224ScoredSep 25, 202649Solar Pro 4UpstageScore−0.254ScoredSep 25, 202650MiniMax M2.7MiniMax229BScore−0.263ScoredSep 15, 202651Gemini 3.5 Flash LiteGoogleScore−0.287ScoredSep 15, 202652Nemotron 3 UltraNVIDIA561BScore−0.319ScoredSep 5, 202653Grok Build 0.1xAIScore−0.392ScoredSep 5, 202654Grok 4.3xAIScore−0.624ScoredSep 5, 202655Gemma 4 31BGoogle31.3BScore−0.715ScoredSep 5, 2026