AI models
GPQA Diamond leaderboard: AI models ranked
195 models ranked. The top score on GPQA Diamond belongs to GPT-6.2 Astra, at 96.8% per Epoch AI, and 27 models are in a statistical tie with it.
GPQA Diamond, on the Epoch AI leaderboard (CC BY 4.0, latest version we checked), is 198 biology, physics and chemistry questions written by PhDs. A quick search almost never settles them. The score is the share of correct answers. Here are the 195 models in this section with a score on GPQA Diamond, per Epoch AI. Rank is shown as a range, because scores within the margin are in a statistical tie.
GPQA Diamond
96.8%
GPT-6.2 Astra
models
195
01
| Model | % | Source |
|---|---|---|
| GPT-6.2 AstraOpenAI · closed | 96.8% between 1st and 4th of 195 top score | Epoch AI |
| Gemini 4 CarbonGoogle · closed | 96.5% between 1st and 5th of 195 statistical tie with the top score | Epoch AI |
| Gemini 4 ArgonGoogle · closed | 96.4% between 1st and 7th of 195 statistical tie with the top score | Epoch AI |
| GPT-6 AstraOpenAI · closed | 95.8% between 1st and 8th of 195 statistical tie with the top score | Epoch AI |
| Claude Sonnet 5.5Anthropic · closed | 95.6% between 2nd and 9th of 195 | Epoch AI |
| Gemini 3.8 FlashGoogle · closed | 95.4% between 3rd and 10th of 195 | Epoch AI |
| GPT-6.1 SolOpenAI · closed | 95.4% between 3rd and 10th of 195 | Epoch AI |
| Gemini 3.7 FlashGoogle · closed | 94.8% between 4th and 19th of 195 | Epoch AI |
| GPT-5.4 ProOpenAI · closed | 94.6% between 5th and 19th of 195 | Epoch AI |
| Gemini 3.1 ProGoogle · closed | 94.4% between 6th and 21st of 195 | Epoch AI |
| GPT-6 SolOpenAI · closed | 94.3% between 8th and 23rd of 195 | Epoch AI |
| GPT-6.1 TerraOpenAI · closed | 94.3% between 8th and 23rd of 195 | Epoch AI |
| Kimi K3.1 ProMoonshot AI · open | 94.2% between 8th and 23rd of 195 | Epoch AI |
| Gemini 3.6 FlashGoogle · closed | 94.1% between 8th and 25th of 195 | Epoch AI |
| GPT-5.5OpenAI · closed | 94% between 8th and 25th of 195 | Epoch AI |
| Grok 4.6xAI · closed | 94% between 8th and 25th of 195 | Epoch AI |
| Claude Fable 5.2Anthropic · closed | 93.9% between 8th and 25th of 195 | Epoch AI |
| Claude Opus 5Anthropic · closed | 93.9% between 8th and 25th of 195 | Epoch AI |
| GPT-5.5 ProOpenAI · closed | 93.9% between 8th and 25th of 195 | Epoch AI |
| GPT-5.6 SolOpenAI · closed | 93.5% between 10th and 28th of 195 | Epoch AI |
| Grok 4.5xAI · closed | 93.4% between 10th and 28th of 195 | Epoch AI |
| GPT-5.4OpenAI · closed | 93.3% between 11th and 29th of 195 | Epoch AI |
| GPT-5.6 TerraOpenAI · closed | 93.3% between 11th and 29th of 195 | Epoch AI |
| Doubao Seed Evolving (Nov 2026)ByteDance · closed | 93.1% between 14th and 29th of 195 | Epoch AI |
| Kimi K3Moonshot AI · open | 93.1% between 14th and 29th of 195 | Epoch AI |
| Gemini 3.5 FlashGoogle · closed | 92.8% between 20th and 30th of 195 | Epoch AI |
| Grok 4.7xAI · closed | 92.7% between 20th and 30th of 195 | Epoch AI |
| Gemini 3 ProGoogle · closed | 92.6% between 20th and 31st of 195 | Epoch AI |
| Qwen3.8 MaxAlibaba · closed | 92.3% between 22nd and 32nd of 195 | Epoch AI |
| GLM-5.2Zhipu AI (Z.ai) · open | 91.9% between 26th and 38th of 195 | Epoch AI |
| GPT-5.6 LunaOpenAI · closed | 91.6% between 28th and 40th of 195 | Epoch AI |
| GPT-5.2OpenAI · closed | 91.4% between 29th and 43rd of 195 | Epoch AI |
| Claude Opus 4.8Anthropic · closed | 91% between 30th and 46th of 195 | Epoch AI |
| DeepSeek V4 Flash 0731DeepSeek · open | 91% between 30th and 46th of 195 | Epoch AI |
| DeepSeek V4 ProDeepSeek · open | 90.9% between 30th and 47th of 195 | Epoch AI |
| GLM-5.3Zhipu AI (Z.ai) · open | 90.9% between 30th and 47th of 195 | Epoch AI |
| MiniMax M3MiniMax · open | 90.9% between 30th and 47th of 195 | Epoch AI |
| Qwen3.7 MaxAlibaba · closed | 90.9% between 30th and 47th of 195 | Epoch AI |
| Kimi K2.6Moonshot AI · open | 90.8% between 31st and 49th of 195 | Epoch AI |
| Claude Opus 5.5Anthropic · closed | 90.6% between 31st and 51st of 195 | Epoch AI |
| Claude Opus 4.6Anthropic · closed | 90.5% between 32nd and 51st of 195 | Epoch AI |
| Claude Sonnet 5Anthropic · closed | 90.5% between 32nd and 51st of 195 | Epoch AI |
| GPT-6 LunaOpenAI · closed | 90.5% between 32nd and 51st of 195 | Epoch AI |
| Claude Opus 4.7Anthropic · closed | 90.2% between 33rd and 54th of 195 | Epoch AI |
| GLM-5.3-FlashZhipu AI (Z.ai) · open | 90.2% between 33rd and 54th of 195 | Epoch AI |
| Mistral Large 4Mistral AI · open | 90.2% between 33rd and 54th of 195 | Epoch AI |
| GLM 5.1Zhipu AI (Z.ai) · open | 89.9% between 35th and 54th of 195 | Epoch AI |
| DeepSeek V4.1 FlashDeepSeek · open | 89.8% between 39th and 55th of 195 | Epoch AI |
| Muse SparkMeta · closed | 89.8% between 39th and 55th of 195 | Epoch AI |
| Claude Haiku 5.5Anthropic · closed | 89.6% between 40th and 56th of 195 | Epoch AI |
| Doubao Seed 2.1 TurboByteDance · closed | 89.6% between 40th and 56th of 195 | Epoch AI |
| Gemini 3 FlashGoogle · closed | 89.4% between 44th and 57th of 195 | Epoch AI |
| Doubao Seed 2.1 ProByteDance · closed | 89.3% between 44th and 57th of 195 | Epoch AI |
| Grok 4.20xAI · closed | 89.3% between 44th and 57th of 195 | Epoch AI |
| Grok 4.3 BetaxAI · closed | 88.8% between 48th and 62nd of 195 | Epoch AI |
| Muse Spark OpenMeta · open | 88.6% between 50th and 64th of 195 | Epoch AI |
| Qwen3.6 PlusAlibaba · closed | 88.4% between 52nd and 66th of 195 | Epoch AI |
| Doubao-Seed-2.0 ProByteDance · closed | 88.2% between 55th and 66th of 195 | Epoch AI |
| Doubao Seed 2.0 Lite (Apr 2026)ByteDance · closed | 88.1% between 55th and 66th of 195 | Epoch AI |
| Kimi K2.7 CodeMoonshot AI · open | 87.9% between 55th and 68th of 195 | Epoch AI |
| Qwen3.7 PlusAlibaba · closed | 87.9% between 55th and 68th of 195 | Epoch AI |
| GLM 5Zhipu AI (Z.ai) · open | 87.8% between 55th and 68th of 195 | Epoch AI |
| GPT-5.1OpenAI · closed | 87.6% between 56th and 68th of 195 | Epoch AI |
| Kimi K2.5Moonshot AI · open | 87.6% between 56th and 68th of 195 | Epoch AI |
| Claude Sonnet 4.6Anthropic · closed | 87.4% between 57th and 69th of 195 | Epoch AI |
| Qwen 3.6 Max (Preview)Alibaba · undisclosed | 87.4% between 57th and 69th of 195 | Epoch AI |
| Grok 4xAI · closed | 87% between 60th and 71st of 195 | Epoch AI |
| GPT-5.4 MiniOpenAI · closed | 86.9% between 60th and 73rd of 195 | Epoch AI |
| Qwen3.5 397B A17BAlibaba · open | 86.4% between 65th and 74th of 195 | Epoch AI |
| GPT-5OpenAI · closed | 86.2% between 67th and 75th of 195 | Epoch AI |
| Claude Opus 4.5Anthropic · closed | 86% between 67th and 75th of 195 | Epoch AI |
| Claude Fable 5Anthropic · closed | 85.9% between 68th and 75th of 195 | Epoch AI |
| Qwen3.6 27BAlibaba · open | 85.9% between 68th and 75th of 195 | Epoch AI |
| Nemotron 3 UltraNVIDIA · open | 85.4% between 69th and 77th of 195 | Epoch AI |
| Gemini 2.5 ProGoogle · closed | 85.3% between 70th and 77th of 195 | Epoch AI |
| Qwen3.5 Plus 2026-02-15Alibaba · undisclosed | 84.8% between 74th and 78th of 195 | Epoch AI |
| Qwen3.6 35B A3BAlibaba · open | 84.8% between 74th and 78th of 195 | Epoch AI |
| Kimi K2 ThinkingMoonshot AI · open | 84.2% between 76th and 83rd of 195 | Epoch AI |
| Qwen3.5-35B-A3BAlibaba · open | 83.5% between 78th and 84th of 195 | Epoch AI |
| DeepSeek V3.2DeepSeek · open | 83.4% between 78th and 84th of 195 | Epoch AI |
| Gemini 3.5 Flash LiteGoogle · closed | 83.3% between 78th and 87th of 195 | Epoch AI |
| GLM 4.7Zhipu AI (Z.ai) · open | 83.3% between 78th and 87th of 195 | Epoch AI |
| Qwen3.6 FlashAlibaba · undisclosed | 83.3% between 78th and 87th of 195 | Epoch AI |
| GPT-5.5 InstantOpenAI · closed | 82.5% between 79th and 89th of 195 | Epoch AI |
| Claude Sonnet 4.5Anthropic · closed | 82.3% between 81st and 89th of 195 | Epoch AI |
| Qwen3.5-FlashAlibaba · undisclosed | 82.3% between 81st and 89th of 195 | Epoch AI |
| Qwen3.7 FlashAlibaba · closed | 82.3% between 81st and 89th of 195 | Epoch AI |
| Gemini 3.1 Flash LiteGoogle · closed | 81.8% between 84th and 89th of 195 | Epoch AI |
| o3OpenAI · closed | 81.8% between 84th and 89th of 195 | Epoch AI |
| Qwen3-235B-A22BAlibaba · open | 80.1% between 90th and 93rd of 195 | Epoch AI |
| Claude 3.7 SonnetAnthropic · closed | 79.7% between 90th and 94th of 195 | Epoch AI |
| o4-miniOpenAI · closed | 79.6% between 90th and 94th of 195 | Epoch AI |
| Claude Sonnet 4Anthropic · closed | 79.2% between 90th and 95th of 195 | Epoch AI |
| Qwen3.5-9BAlibaba · open | 79% between 91st and 95th of 195 | Epoch AI |
| GPT-5.4 NanoOpenAI · closed | 78.5% between 93rd and 96th of 195 | Epoch AI |
| Doubao-Seed-1.6ByteDance · closed | 77.6% between 95th and 99th of 195 | Epoch AI |
| Claude Opus 4.1Anthropic · closed | 77.3% between 96th and 102nd of 195 | Epoch AI |
| o3-miniOpenAI · closed | 77% between 96th and 102nd of 195 | Epoch AI |
| o1OpenAI · closed | 76.8% between 96th and 104th of 195 | Epoch AI |
| Claude Opus 4Anthropic · closed | 76.3% between 97th and 104th of 195 | Epoch AI |
| DeepSeek R1DeepSeek · open | 76.3% between 97th and 104th of 195 | Epoch AI |
| Grok-3 minixAI · closed | 76.3% between 97th and 104th of 195 | Epoch AI |
| gpt-oss-120bOpenAI · open | 75.8% between 99th and 105th of 195 | Epoch AI |
| Grok 3xAI · closed | 75.8% between 99th and 105th of 195 | Epoch AI |
| GPT-5 MiniOpenAI · closed | 75% between 103rd and 105th of 195 | Epoch AI |
| Gemma 4 26B A4BGoogle · open | 73.2% between 106th and 107th of 195 | Epoch AI |
| Qwen3-MaxAlibaba · closed | 72.6% between 106th and 107th of 195 | Epoch AI |
| Claude Haiku 4.5Anthropic · closed | 71.2% 108th of 195 | Epoch AI |
| GPT-5 NanoOpenAI · closed | 69.4% between 109th and 110th of 195 | Epoch AI |
| GPT-4.5OpenAI · closed | 68.7% between 109th and 110th of 195 | Epoch AI |
| DeepSeek V3DeepSeek · open | 67.6% between 111th and 113th of 195 | Epoch AI |
| Llama 4 MaverickMeta · open | 67% between 111th and 113th of 195 | Epoch AI |
| GPT-4.1OpenAI · closed | 66.9% between 111th and 113th of 195 | Epoch AI |
| GPT-4.1 MiniOpenAI · closed | 65.8% between 114th and 118th of 195 | Epoch AI |
| Gemini 2.0 ProGoogle · closed | 65.7% between 114th and 118th of 195 | Epoch AI |
| Qwen3 32BAlibaba · open | 65.7% between 114th and 118th of 195 | Epoch AI |
| QWQ-PlusAlibaba · closed | 65.4% between 114th and 118th of 195 | Epoch AI |
| QwQ-32BAlibaba · open | 65.3% between 114th and 118th of 195 | Epoch AI |
| DeepSeek-R1-Distill-Qwen-32BDeepSeek · open | 64.1% between 119th and 121st of 195 | Epoch AI |
| Gemini 2.0 FlashGoogle · closed | 64.1% between 119th and 121st of 195 | Epoch AI |
| Qwen3 14BAlibaba · open | 63.8% between 119th and 121st of 195 | Epoch AI |
| o1-miniOpenAI · closed | 62.4% between 122nd and 123rd of 195 | Epoch AI |
| Qwen3 30B A3BAlibaba · open | 61.7% between 122nd and 124th of 195 | Epoch AI |
| gpt-oss-20bOpenAI · open | 60.8% between 123rd and 124th of 195 | Epoch AI |
| Mistral Medium 3Mistral AI · closed | 59.5% 125th of 195 | Epoch AI |
| Gemini 1.5 ProGoogle · closed | 57.2% between 126th and 128th of 195 | Epoch AI |
| Gemini 2.0 Flash ThinkingGoogle · closed | 57.1% between 126th and 131st of 195 | Epoch AI |
| Qwen3-8BAlibaba · open | 56.8% between 126th and 131st of 195 | Epoch AI |
| Magistral Small 1.0Mistral AI · open | 56.1% between 127th and 133rd of 195 | Epoch AI |
| Phi 4Microsoft · open | 56.1% between 127th and 133rd of 195 | Epoch AI |
| Qwen2.5-MaxAlibaba · closed | 56.1% between 127th and 133rd of 195 | Epoch AI |
| DeepSeek-R1-Distill-Llama-70BDeepSeek · open | 55.7% between 129th and 133rd of 195 | Epoch AI |
| Claude 3.5 Sonnet (October 2024)Anthropic · closed | 55.3% between 129th and 133rd of 195 | Epoch AI |
| Claude 3.5 SonnetAnthropic · closed | 54% between 134th and 135th of 195 | Epoch AI |
| Grok-2xAI · closed | 53.8% between 134th and 135th of 195 | Epoch AI |
| Qwen3-4BAlibaba · open | 52.3% between 136th and 138th of 195 | Epoch AI |
| Llama 4 ScoutMeta · open | 51.8% between 136th and 139th of 195 | Epoch AI |
| Mistral Large 2Mistral AI · open | 51.3% between 136th and 140th of 195 | Epoch AI |
| Llama 3.1-405BMeta · open | 50.9% between 137th and 140th of 195 | Epoch AI |
| o1-previewOpenAI · closed | 50.3% between 138th and 140th of 195 | Epoch AI |
| Mistral Small 3.2Mistral AI · open | 49.1% between 141st and 144th of 195 | Epoch AI |
| Qwen2.5-72BAlibaba · open | 49.1% between 141st and 144th of 195 | Epoch AI |
| GPT-4.1 NanoOpenAI · closed | 48.9% between 141st and 144th of 195 | Epoch AI |
| Qwen-PlusAlibaba · closed | 48.1% between 141st and 151st of 195 | Epoch AI |
| Gemma 3 27BGoogle · open | 47.7% between 144th and 151st of 195 | Epoch AI |
| Magistral Small 1.2Mistral AI · open | 47.6% between 144th and 152nd of 195 | Epoch AI |
| Mistral Small 3.1Mistral AI · open | 47.5% between 144th and 152nd of 195 | Epoch AI |
| Llama 3.3 70BMeta · open | 47.4% between 144th and 152nd of 195 | Epoch AI |
| Gemini 1.5 FlashGoogle · closed | 47.3% between 144th and 153rd of 195 | Epoch AI |
| Mistral Small 3Mistral AI · open | 47.3% between 144th and 153rd of 195 | Epoch AI |
| Claude 3 OpusAnthropic · closed | 47.2% between 144th and 153rd of 195 | Epoch AI |
| GPT-4 Turbo (Apr 2024)OpenAI · closed | 46.6% between 146th and 154th of 195 | Epoch AI |
| Tulu 3 (Tülu 3) 70BAi2 (Allen Institute) · open | 46.3% between 149th and 154th of 195 | Epoch AI |
| Qwen2.5-32BAlibaba · open | 46.1% between 152nd and 154th of 195 | Epoch AI |
| DeepSeek-R1-Distill-Qwen-14BDeepSeek · open | 44.7% between 155th and 156th of 195 | Epoch AI |
| Llama 3.1-70BMeta · open | 44.2% between 155th and 157th of 195 | Epoch AI |
| WizardLM-2 8x22BMicrosoft · open | 43.4% between 156th and 158th of 195 | Epoch AI |
| GPT-4 Turbo (Nov 2023)OpenAI · closed | 42.4% between 157th and 159th of 195 | Epoch AI |
| Qwen-TurboAlibaba · closed | 41.8% between 158th and 161st of 195 | Epoch AI |
| Llama 3.2 90BMeta · open | 41% between 159th and 163rd of 195 | Epoch AI |
| Qwen2-72BAlibaba · open | 40.8% between 159th and 163rd of 195 | Epoch AI |
| Claude 3 SonnetAnthropic · closed | 40.6% between 160th and 163rd of 195 | Epoch AI |
| Llama 3-70BMeta · open | 40.6% between 160th and 163rd of 195 | Epoch AI |
| Gemma 3 12BGoogle · open | 39.5% between 164th and 165th of 195 | Epoch AI |
| Mistral LargeMistral AI · closed | 38.8% between 164th and 167th of 195 | Epoch AI |
| Claude 3.5 HaikuAnthropic · closed | 38.1% between 165th and 168th of 195 | Epoch AI |
| Qwen3-1.7BAlibaba · open | 38% between 165th and 168th of 195 | Epoch AI |
| GPT-4o-miniOpenAI · closed | 37.7% between 166th and 168th of 195 | Epoch AI |
| Gemma 2 27BGoogle · open | 36.5% between 169th and 171st of 195 | Epoch AI |
| Claude 3 HaikuAnthropic · closed | 36.3% between 169th and 171st of 195 | Epoch AI |
| Qwen2.5-7BAlibaba · open | 35.5% between 169th and 172nd of 195 | Epoch AI |
| Claude 2Anthropic · closed | 34.7% between 171st and 174th of 195 | Epoch AI |
| Mixtral 8x22BMistral AI · open | 34.1% between 172nd and 175th of 195 | Epoch AI |
| Gemini 1.0 ProGoogle · closed | 34% between 172nd and 177th of 195 | Epoch AI |
| DeepSeek-R1-Distill-Qwen-1.5BDeepSeek · open | 33.6% between 173rd and 177th of 195 | Epoch AI |
| Claude 2.1Anthropic · closed | 33% between 174th and 178th of 195 | Epoch AI |
| Gemini 1.5 Flash 8BGoogle · closed | 33% between 174th and 178th of 195 | Epoch AI |
| Yi-1.5-34B01.AI · open | 32% between 176th and 178th of 195 | Epoch AI |
| Qwen1.5-32BAlibaba · open | 30.7% between 179th and 181st of 195 | Epoch AI |
| Mixtral 8x7BMistral AI · open | 30.6% between 179th and 181st of 195 | Epoch AI |
| Mistral NemoMistral AI · open | 29.9% between 179th and 181st of 195 | Epoch AI |
| Qwen1.5-72BAlibaba · open | 28.8% 182nd of 195 | Epoch AI |
| phi-3-medium 14BMicrosoft · open | 27.6% between 183rd and 186th of 195 | Epoch AI |
| Gemma 2 9BGoogle · open | 27.5% between 183rd and 186th of 195 | Epoch AI |
| Ministral 8BMistral AI · open | 27.1% between 183rd and 188th of 195 | Epoch AI |
| Llama 3.1-8BMeta · open | 27% between 183rd and 188th of 195 | Epoch AI |
| Llama 2-70BMeta · open | 26.3% between 185th and 189th of 195 | Epoch AI |
| Llama 3-8BMeta · open | 26.1% between 185th and 189th of 195 | Epoch AI |
| Ministral 3BMistral AI · closed | 25.3% between 187th and 190th of 195 | Epoch AI |
| DeepSeek LLM 67BDeepSeek · open | 24.6% between 189th and 191st of 195 | Epoch AI |
| Llama 3.2 1BMeta · open | 23.9% between 190th and 192nd of 195 | Epoch AI |
| Gemma 3 4BGoogle · open | 23.2% between 191st and 192nd of 195 | Epoch AI |
| Gemma 3 1BGoogle · open | 19.9% 193rd of 195 | Epoch AI |
| Mistral 7BMistral AI · open | 15.2% between 194th and 195th of 195 | Epoch AI |
| Yi-34B01.AI · open | 14.7% between 194th and 195th of 195 | Epoch AI |
| Gemini 4 ArgonGoogle · closed | 96.5%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemini 3.8 FlashGoogle · closed | 95.6%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-6 Astra ProOpenAI · undisclosed | 95.5%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemini 3.1 ProGoogle · closed | 95.2%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-6.1 SolOpenAI · closed | 94.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Ling 3.1 FlashAnt Group · undisclosed | 94.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-6 AstraOpenAI · closed | 94.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemini 3.7 FlashGoogle · closed | 94.3%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-5.6 Sol ProOpenAI · undisclosed | 94.1%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-5.5OpenAI · closed | 93.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Kimi K3.1 ProMoonshot AI · open | 93.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Kimi K3Moonshot AI · open | 93.5%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemini 3.6 FlashGoogle · closed | 93.4%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Claude Sonnet 5.5Anthropic · closed | 93%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| DeepSeek V4.1 FlashDeepSeek · open | 92.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-5.6 SolOpenAI · closed | 92.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Grok 4.6xAI · closed | 92.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemini 3.5 FlashGoogle · closed | 92.6%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GLM-5.3-FlashZhipu AI (Z.ai) · open | 92.5%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-6 SolOpenAI · closed | 92.5%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Nemotron 3.5 Super VLNVIDIA · open | 92.2%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| DeepSeek V4 ProDeepSeek · open | 91.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-6.1 TerraOpenAI · closed | 91.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Grok 4.7xAI · closed | 91.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3.8 2.4T A95BAlibaba · open | 91.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| MiniMax M3MiniMax · open | 91.3%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-5.6 Luna ProOpenAI · undisclosed | 91%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Muse Spark OpenMeta · open | 91%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| DeepSeek V4 Flash Vision ExpDeepSeek · open | 90.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Kimi K2.7 CodeMoonshot AI · open | 90.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3.7 MaxAlibaba · closed | 90.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Hy4Tencent · open | 90.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| MiMo V2.6 ProXiaomi · open | 90.6%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3.8 FlashAlibaba · open | 90.5%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemini 3 Flash PreviewGoogle · undisclosed | 90.4%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-5.4OpenAI · closed | 90.3%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Claude Opus 5.5Anthropic · closed | 90.2%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Kimi K2.5Moonshot AI · open | 90.2%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Claude Opus 4.8Anthropic · closed | 90.1%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| DeepSeek V4 Flash 0423DeepSeek · open | 90.1%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Claude Fable 5.1Anthropic · closed | 89.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Claude Opus 5Anthropic · closed | 89.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Claude Opus 4.7Anthropic · closed | 89.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GLM-5.2Zhipu AI (Z.ai) · open | 89.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Kimi K2.6Moonshot AI · open | 89.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-5.6 TerraOpenAI · closed | 89.3%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| MiniMax M3.1 FlashMiniMax · open | 89.3%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-6 Luna ProOpenAI · undisclosed | 89%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Nova Micro 1.0Amazon · undisclosed | 88.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-5.6 LunaOpenAI · closed | 88.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-6 LunaOpenAI · closed | 88.2%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3.5 397B A17BAlibaba · open | 88.1%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GLM-5.3Zhipu AI (Z.ai) · open | 87.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3.7 PlusAlibaba · closed | 87.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Claude Fable 5Anthropic · closed | 87.4%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3.8 27BAlibaba · open | 87.2%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Claude Haiku 5.5Anthropic · closed | 87.1%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-5.1OpenAI · closed | 87%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Claude Sonnet 5Anthropic · closed | 86.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| MiMo-V2.5-ProXiaomi · open | 86.5%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Claude Opus 4.5Anthropic · closed | 86.4%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Claude Opus 4.6Anthropic · closed | 86.4%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| MiniMax M2.7MiniMax · open | 86.1%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemma 4 31BGoogle · open | 85.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-5OpenAI · closed | 85.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| MiMo-V2.6-FlashXiaomi · open | 85.2%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3.6 27BAlibaba · open | 85.1%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3.6 35B A3BAlibaba · open | 85.1%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| MiniMax M2.5MiniMax · open | 85%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| MiMo-V2.5Xiaomi · open | 84.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Ling 3.0 Flash VLAnt Group · open | 84.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3.5-122B-A10BAlibaba · open | 84.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemini 3.5 Flash LiteGoogle · closed | 84.4%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GLM 4.6Zhipu AI (Z.ai) · open | 84.3%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-5.4 MiniOpenAI · closed | 84.2%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GLM 5.1Zhipu AI (Z.ai) · open | 84.1%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GLM 4.7Zhipu AI (Z.ai) · open | 83.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| DeepSeek V3.2DeepSeek · open | 83%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3.5-35B-A3BAlibaba · open | 82.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemini 2.5 ProGoogle · closed | 82.6%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| DeepSeek V3.1 TerminusDeepSeek · open | 82.4%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemini 3.1 Flash LiteGoogle · closed | 82.1%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Muse Glimmer 30BMeta · open | 82.1%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Claude Sonnet 4.5Anthropic · closed | 81.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| DeepSeek V3.1DeepSeek · open | 81.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GLM 5Zhipu AI (Z.ai) · open | 81%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-5 MiniOpenAI · closed | 81%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3.5-9BAlibaba · open | 81%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| DeepSeek R1DeepSeek · open | 79.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3 235B A22B Thinking 2507Alibaba · open | 79.1%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemma 4 26B A4BGoogle · open | 78.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-5.4 NanoOpenAI · closed | 77.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| gpt-oss-120bOpenAI · open | 77.4%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3-235B-A22BAlibaba · open | 77.4%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemini 2.5 FlashGoogle · closed | 76.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Ling 3.0 FlashAnt Group · open | 76.5%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3 Coder NextAlibaba · open | 76.3%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Mistral Small 4Mistral AI · open | 75.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Nemotron 3.5 LightningNVIDIA · open | 74.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Claude Haiku 4.5Anthropic · closed | 73%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3 Next 80B A3B InstructAlibaba · open | 71.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3 VL 235B A22B InstructAlibaba · open | 71.1%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Claude Sonnet 4.6Anthropic · closed | 70.6%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-5 NanoOpenAI · closed | 70.5%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Nemotron 3 Nano 30B A3BNVIDIA · open | 68.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| gpt-oss-20bOpenAI · open | 68.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GLM 4.6VZhipu AI (Z.ai) · open | 66.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3 VL 30B A3B InstructAlibaba · open | 66.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| DeepSeek V3DeepSeek · open | 66.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Llama 4 MaverickMeta · open | 66.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-4.1OpenAI · closed | 66%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Granite 4.2 8BIBM · open | 65.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-4.1 MiniOpenAI · closed | 65.2%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3 30B A3B Instruct 2507Alibaba · open | 64.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3 32BAlibaba · open | 63.3%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3 14BAlibaba · open | 61.2%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3-CoderAlibaba · open | 61.2%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3 30B A3BAlibaba · open | 60.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemini 2.5 Flash LiteGoogle · undisclosed | 59.6%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GLM 4.7 FlashZhipu AI (Z.ai) · open | 58%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3 VL 8B InstructAlibaba · open | 53.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen3 Coder 30B A3B InstructAlibaba · open | 53.4%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-4oOpenAI · closed | 52.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-4.1 NanoOpenAI · closed | 50.5%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Llama 3.3 70B InstructMeta · open | 49.8%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen2.5 72B InstructAlibaba · open | 47.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Qwen2.5 VL 72B InstructAlibaba · open | 45.7%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| GPT-4o-miniOpenAI · closed | 43.6%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemma 3 12BGoogle · open | 40.6%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Hy-MT2-30B-A3BTencent · open | 38.9%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Llama 3.1 8B InstructMeta · open | 29%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Llama 3.2 3B InstructMeta · open | 28.6%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Gemma 3 4BGoogle · open | 27.1%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
| Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)Google · undisclosed | 11.6%, per OpenRouter Benchmarks | OpenRouter Benchmarks |
FAQ
Which model has the top score on GPQA Diamond?
The top score on GPQA Diamond belongs to GPT-6.2 Astra, at 96.8% per Epoch AI, and 27 models are in a statistical tie with it. Scores from another source show that source on their own row, with no badge and no statistical tie with the top score from Epoch AI.
What does GPQA Diamond measure?
GPQA Diamond, on the Epoch AI leaderboard (CC BY 4.0, latest version we checked), is 198 biology, physics and chemistry questions written by PhDs. A quick search almost never settles them. The score is the share of correct answers.
What does “statistical tie with the top score” mean?
It means the gap to the top score is within the 95% margin, which is never below 1 point. On GPQA Diamond, near 90%, that margin is about 5.9 points. That is why rank is shown as a range, «between Xth and Yth of N»: among tied models, the order may be measurement noise.
Which models are in this ranking?
The 195 models in this section with a published score on GPQA Diamond, per Epoch AI, plus, on separate rows, scores from another source with that source shown. No version inherits another's score.
How to read
- Benchmark scores are percentages, and higher is better.
- Rank is shown as a range, «between 1st and 28th of 195». Within the range, the gap to neighboring models falls within the benchmark's margin. The «top score» badge marks the model with the benchmark's top score, and the «statistical tie with the top score» badge marks models whose gap to it is within the margin.
- On a saturated benchmark, the table shows the scores with the badge and without a rank order.
Sources
- Epoch AI, CC BY 4.0, latest version we checked.
- OpenRouter Benchmarks, CC BY 4.0, latest version we checked.
- OpenRouter, dados públicos da API; exports de benchmarks e rankings em CC BY 4.0, latest version we checked.
Dates come from the sources: OpenRouter listing or Hugging Face repo creation, which may differ from the official launch. Family and version were grouped by rule from the name.
How to cite
If you use this data, cite the page and the date you accessed it.
Adriano Meirinho. "GPQA Diamond leaderboard: AI models ranked". meirinho.com.br. https://meirinho.com.br/en/ia/modelos/benchmark/gpqa-diamond/. Accessed November 21, 2026.
Third-party data follows the license of each source. The full table is available as CSV and JSON.