Benchmark Registry
2,832 benchmark results
110 models84 benchmarks217 versions
Explore Benchmarks
Browse evaluations across models
- SWE-bench98 models evaluated299 results
- GPQA95 models evaluated304 results
- MMLU-Pro91 models evaluated141 results
- Terminal-Bench86 models evaluated275 results
- Humanity's Last Exam68 models evaluated210 results
Latest Additions
New benchmark results added to the Registry
All Models
Alphabetical directory
C
- Claude 3.7 SonnetAnthropicGPQA Diamond — Epoch AI harness66.0%
- Claude Fable 5AnthropicAPEX-Agents Original (max)63.6%
- Claude Fable 5.1AnthropicAPEX-Agents Original (high)59.7%
- Claude Haiku 4.5AnthropicAutomationBench 1.0.60.5%
- Claude Haiku 5.5AnthropicCursorBench 4.0 (high)42.3%
- Claude Opus 4AnthropicGPQA Diamond — Epoch AI harness69.2%
- Claude Opus 4.1AnthropicGPQA Diamond — Epoch AI harness73.2%
- Claude Opus 4.5AnthropicGPQA Diamond — Epoch AI harness80.7%
- Claude Opus 4.6AnthropicAPEX-Agents Original (max)46.3%
- Claude Opus 4.7AnthropicAPEX-Agents Original (max)49.2%
- Claude Opus 4.8AnthropicAPEX-Agents Original (max)48.9%
- Claude Opus 5AnthropicAPEX-Agents Original (max)65.8%
- Claude Opus 5.5AnthropicAPEX-Agents Original (max)73.5%
- Claude Sonnet 4AnthropicGPQA Diamond — Epoch AI harness66.7%
- Claude Sonnet 4.5AnthropicGPQA Diamond — Epoch AI harness73.7%
- Claude Sonnet 4.6AnthropicAPEX-Agents Original (high)43.0%
- Claude Sonnet 5AnthropicAPEX-Agents Original (max)54.5%
- Claude Sonnet 5.5AnthropicAPEX-Agents Original (max)75.5%
- Composer 1CursorCursorBench 3.038.0%
- Composer 1.5CursorCursorBench 3.044.2%
- Composer 2CursorCursorBench 3.061.3%
- Composer 2.5CursorVibe Code Bench 1.1 — Cursor CLI, October 2026 evaluation49.6%
D
- DeepSeek-V3.2DeepSeek AIAPEX-Agents Original21.3%
- DeepSeek-V4-FlashDeepSeek AIAutomationBench 1.0.6 (max)18.1%
- DeepSeek-V4-ProDeepSeek AIBrowseComp 130-question subset — Mercor web research agent (max)62.3%
- Devstral 2Mistral AISWE-bench Verified — Vals AI mini-swe-agent62.8%
- Devstral Small 2Mistral AISWE-bench Multilingual55.7%
G
- Gemini 2.0 FlashGoogle (DeepMind)GPQA Diamond — Epoch AI harness64.1%
- Gemini 2.5 FlashGoogle (DeepMind)Humanity's Last Exam Text only — Scale AI evaluation12.6%
- Gemini 2.5 ProGoogle (DeepMind)GPQA Diamond — Epoch AI harness83.8%
- Gemini 3 FlashGoogle (DeepMind)GPQA Diamond — Epoch AI harness83.2%
- Gemini 3 ProGoogle (DeepMind)GPQA Diamond — Epoch AI harness92.6%
- Gemini 3.1 ProGoogle (DeepMind)APEX-Agents Original (high)35.3%
- Gemini 3.5 FlashGoogle (DeepMind)APEX-Agents Original (high)27.5%
- Gemini 3.6 FlashGoogle (DeepMind)APEX-Agents Original (high)46.9%
- Gemini 3.7 FlashGoogle (DeepMind)APEX-Agents Original (high)67.8%
- Gemini 3.8 FlashGoogle (DeepMind)APEX-Agents Original (high)64.3%
- Gemini 4 ArgonGoogle (DeepMind)APEX-Agents Original (high)82.4%
- Gemma 4 12B InstructGoogle (DeepMind)BIG-Bench Extra Hard Original — micro average (thinking mode)53.0%
- Gemma 4 26B A4B InstructGoogle (DeepMind)BIG-Bench Extra Hard Original — micro average (thinking mode)64.8%
- Gemma 4 31B InstructGoogle (DeepMind)AutomationBench 1.0.6 (max)1.7%
- Gemma 4 E2B InstructGoogle (DeepMind)BIG-Bench Extra Hard Original — micro average (thinking mode)21.9%
- Gemma 4 E4B InstructGoogle (DeepMind)BIG-Bench Extra Hard Original — micro average (thinking mode)33.1%
- GLM-4.5Z.aiHumanity's Last Exam Text only — Scale AI evaluation9.6%
- GLM-4.5-AirZ.aiHumanity's Last Exam Text only — Scale AI evaluation9.4%
- GLM-4.6Z.aiSWE-bench Pro Public — Scale AI SWE-Agent9.7%
- GLM-4.7Z.aiGPQA Diamond — Epoch AI harness83.3%
- GLM-5Z.aiDeepSWE 1.1 — Mercor mini-swe-agent, 500 steps2.1%
- GLM-5.1Z.aiAPEX-Agents Original40.9%
- GLM-5.2Z.aiAPEX-Agents Original (max)45.2%
- GLM-5.3Z.aiAPEX-Agents Original (max)56.6%
- GLM-5.3-FlashZ.aiAPEX-Agents Original (max)52.8%
- GPT-4.1OpenAIGPQA Diamond — Epoch AI harness66.9%
- GPT-4.1 miniOpenAIGPQA Diamond — Epoch AI harness65.8%
- GPT-4.1 nanoOpenAIGPQA Diamond — Epoch AI harness48.9%
- GPT-4.5 PreviewOpenAIGPQA Diamond — Epoch AI harness68.7%
- GPT-5OpenAIGPQA Diamond — Epoch AI harness (high)86.2%
- GPT-5 miniOpenAIGPQA Diamond — Epoch AI harness (high)75.0%
- GPT-5 nanoOpenAIGPQA Diamond — Epoch AI harness (high)69.4%
- GPT-5.2OpenAIGPQA Diamond — Epoch AI harness (high)88.2%
- GPT-5.3-CodexOpenAISWE Atlas Codebase QnA — Public; Codex (xHigh)32.6%
- GPT-5.4OpenAIAPEX-Agents Original (xhigh)52.4%
- GPT-5.4 miniOpenAIGPQA Diamond — Epoch AI harness (high)83.6%
- GPT-5.4 nanoOpenAIGPQA Diamond — Epoch AI harness (high)78.5%
- GPT-5.5OpenAIAPEX-Agents Original (xhigh)55.1%
- GPT-5.6 LunaOpenAIAPEX-Agents Original (max)43.0%
- GPT-5.6 SolOpenAIAutomationBench 1.0.63.8%
- GPT-5.6 TerraOpenAIAPEX-Agents Original (max)58.2%
- GPT-6 AstraOpenAIAPEX-Agents Original (max)64.7%
- GPT-6 LunaOpenAIAPEX-Agents Original (max)44.3%
- GPT-6 SolOpenAIAPEX-Agents Original (max)54.3%
- GPT-6.1 SolOpenAIAPEX-Agents Original (max)60.0%
- gpt-oss-120bOpenAIAPEX-Agents Original (high)4.4%
- Grok 4SpaceXAIGPQA Diamond — Epoch AI harness87.0%
- Grok 4.5SpaceXAIAPEX-Agents Original (high)56.2%
- Grok 4.6SpaceXAIAPEX-Agents Original (xhigh)65.3%
- Grok 4.7SpaceXAIAPEX-Agents Original (xhigh)54.6%
I
- InklingThinking Machines LabAPEX-Agents Original (high)33.8%
- Inkling-SmallThinking Machines LabChartography October 2026 leaderboard (xHigh)9.1%
K
M
- MAI-Code-1-FlashMicrosoft AIGPQA Diamond84.6%
- MAI-Thinking-1Microsoft AIRiemann-bench October 2026 leaderboard4.8%
- MiniMax M2MiniMaxSWE-bench Verified — mini-swe-agent61.0%
- MiniMax M2.5MiniMaxSWE Atlas Codebase QnA — Public; mini-swe-agent10.3%
- MiniMax M2.7MiniMaxAutomationBench 1.0.6 (high)0.5%
- MiniMax M3MiniMaxAPEX-Agents Original (high)37.7%
- Mistral Large 3Mistral AIChartography October 2026 leaderboard9.0%
- Mistral Medium 3.5Mistral AIFinance Agent Benchmark 2 — October 2026 evaluation (high)32.1%
- Mistral Small 4Mistral AIVibe Code Bench 1.1 — OpenHands, October 2026 evaluation (high)0.0%
- Muse Glimmer 30BMeta AI (originally Facebook AI Research)Chartography October 2026 leaderboard (xHigh)17.2%
- Muse SparkMeta AI (originally Facebook AI Research)GPQA Diamond — Epoch AI harness89.8%
- Muse Spark 1.1Meta AI (originally Facebook AI Research)APEX-Agents Original (xhigh)31.8%
- Muse Spark 1.2Meta AI (originally Facebook AI Research)APEX-Agents Original (xhigh)36.4%
- Muse Spark 1.3Meta AI (originally Facebook AI Research)APEX-Agents Original (max)47.6%
N
O
P
Q
- Qwen3-Max-InstructAlibaba (Tongyi)SWE-bench Verified69.6%
- Qwen3-Max-ThinkingAlibaba (Tongyi)GPQA Diamond — Vals AI harness84.8%
- Qwen3.5-397B-A17BAlibaba (Tongyi)APEX-Agents Original24.9%
- Qwen3.6-35B-A3BAlibaba (Tongyi)GPQA Diamond — Epoch AI harness83.8%
- Qwen3.6-PlusAlibaba (Tongyi)AutomationBench 1.0.6 (high)4.3%
- Qwen3.7-MaxAlibaba (Tongyi)GPQA Diamond — Epoch AI harness90.9%
- Qwen3.7-PlusAlibaba (Tongyi)AutomationBench 1.0.63.7%
- Qwen3.8-MaxAlibaba (Tongyi)APEX-Agents Original (xhigh)63.3%