GPQA Diamond leaderboard
- models measured
- 91
- lead over #2
- +0.2 pts
- median
- 84.8%
Leader on GPQA Diamond · read October 22
On October 22, 2026, 95 of our catalog's models had a GPQA Diamond result, and Google Gemini 3.1 Pro Preview led with 95.5%, ahead of OpenAI GPT-5.6 Sol at 95.2%.
GPQA asks graduate-level multiple-choice questions in biology, physics and chemistry, written by domain experts to be hard to answer by searching the web. The number is the share answered correctly, as run by Vals AI.
Role in the index
Limit: Vals AI calls this board largely saturated: when we read it on October 15, 2026, 24 of the 136 models it tested scored 90% or higher, so the top of the board separates models less than the bottom does.
GPQA Diamond is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Latency, as printed | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|
| 1 | Gemini 3.1 Pro PreviewGoogle | 95.5% | 1m 06s | 32.3 | $12.00 | |
| 2 | GPT-5.6 SolOpenAI | 95.2% | 58 s | 73.5 | $20.00 | |
| 3 | Grok 4.6xAI | 94.7% | 3m 29s | 65.7 | $6.00 | |
| 4 | Gemini 3.8 FlashGoogle | 94.4% | 20 s | 61.6 | $3.75 | |
| 5 | Gemini 3.7 FlashGoogle | 93.9% | 8.8 s | 58.0 | $3.75 | |
| 6 | Qwen3.8 Max (0902)Qwen | 93.7% | 4m 27s | 62.4 | $6.00 | |
| 7 | Claude Fable 5.1Anthropic | 93.4% | 46 s | 84.2 | $50.00 | |
| 8 | Claude Opus 5Anthropic | 93.4% | 40 s | 83.0 | $25.00 | |
| 9 | Gemini 3.6 FlashGoogle | 93.4% | 18 s | 62.5 | $3.75 | |
| 10 | Claude Fable 5Anthropic | 93.2% | 1m 40s | 79.3 | $50.00 | |
| 11 | GPT-5.5OpenAI | 93.2% | 1m 52s | 71.3 | $30.00 | |
| 12 | Kimi K3Moonshot AI | 92.9% | 2m 02s | 62.3 | $15.00 | |
| 13 | Grok 4.5xAI | 92.9% | 3m 38s | 58.6 | $6.00 | |
| 14 | Gemini 3.5 FlashGoogle | 92.7% | 28 s | 39.8 | $9.00 | |
| 15 | MiniMax M3MiniMax | 92.7% | 4m 39s | 39.2 | $1.20 | |
| 16 | Claude Opus 4.8Anthropic | 92.4% | 2m 14s | 65.0 | $25.00 | |
| 17 | DeepSeek V4 Pro 0813DeepSeek | 92.4% | 4m 39s | 56.1 | $1.98 | |
| 18 | GPT-5.2OpenAI | 91.7% | 2m 53s | 55.9 | $14.00 | |
| 19 | GPT-5.4OpenAI | 91.7% | 3m 23s | 64.4 | $15.00 | |
| 20 | GPT-5.6 LunaOpenAI | 91.7% | 54 s | 58.0 | $1.20 | |
| 21 | Grok 4.3xAI | 91.4% | 12m 20s | 37.4 | $2.50 | |
| 22 | Muse Spark 1.1Meta | 91.2% | 1m 15s | 70.3 | $4.25 | |
| 23 | GPT-5.6 TerraOpenAI | 90.9% | 37 s | 63.9 | $12.00 | |
| 24 | Claude Opus 4.7Anthropic | 90.2% | 3m 23s | 67.5 | $25.00 | |
| 25 | Qwen3.7 MaxQwen | 90.2% | 2m 07s | 54.1 | $4.425 | |
| 26 | DeepSeek V4 Flash 0731DeepSeek | 89.9% | 1m 20s | 51.8 | $0.396 | |
| 27 | Claude Opus 4.6Anthropic | 89.7% | 5m 19s | 57.2 | $25.00 | |
| 28 | DeepSeek V4 Pro 0423DeepSeek | 89.4% | 15m 00s | 38.7 | $2.976 | |
| 29 | Kimi K2.6Moonshot AI | 89.1% | 9m 05s | 52.1 | $4.00 | |
| 30 | Claude Sonnet 5Anthropic | 88.9% | 1m 03s | 57.7 | $10.00 | |
| 31 | Qwen3.8 27BQwen | 88.9% | 1m 56s | 74.2 | $2.55 | |
| 32 | Grok 4.20xAI | 88.6% | 59 s | 33.3 | $2.50 | |
| 33 | GLM 5.3Z.ai | 88.1% | 3m 31s | 70.6 | $4.40 | |
| 34 | Gemini 3 Flash PreviewGoogle | 87.9% | 1m 24s | 43.7 | $3.00 | |
| 35 | Qwen3.5 Plus 2026-04-20Qwen | 87.4% | 1m 44s | 39.5 | $1.80 | |
| 36 | Qwen3.6 PlusQwen | 87.4% | 2m 02s | 44.1 | $1.95 | |
| 37 | T InklingThinking Machines | 87.1% | 5m 09s | 31.5 | $4.05 | |
| 38 | MiniMax M2.7MiniMax | 86.6% | 2m 52s | 36.2 | $1.20 | |
| 39 | GPT-5.1OpenAI | 86.6% | 1m 26s | 40.4 | $10.00 | |
| 40 | GLM 5.3 FlashZ.ai | 86.4% | 2m 11s | 57.7 | $0.50 | |
| 41 | Nemotron 3 UltraNVIDIA | 86.1% | 2m 38s | 30.0 | $2.40 | |
| 42 | Claude Opus 4.5Anthropic | 85.9% | 2m 19s | 44.0 | $25.00 | |
| 43 | Claude Sonnet 4.6Anthropic | 85.6% | 6m 18s | 57.3 | $15.00 | |
| 44 | GPT-5OpenAI | 85.6% | 2m 50s | 38.8 | $10.00 | |
| 45 | GLM 5.2Z.ai | 85.6% | 3m 45s | 61.1 | $4.40 | |
| 46 | Ling 3.0 FlashinclusionAI | 84.8% | 44 s | 29.6 | $0.1215 | |
| 47 | GLM 5.1Z.ai | 84.5% | 4m 10s | 49.2 | $4.40 | |
| 48 | Kimi K2.5Moonshot AI | 84.1% | 2m 32s | 38.6 | $2.85 | |
| 49 | o3OpenAI | 84.1% | 1m 06s | Not benchmarked | $8.00 | |
| 50 | Gemini 3.5 Flash LiteGoogle | 83.8% | 14 s | 49.0 | $2.50 | |
| 51 | T Inkling SmallThinking Machines | 83.6% | 4m 20s | 27.3 | $1.20 | |
| 52 | GLM 5Z.ai | 83.3% | 4m 03s | 41.7 | $3.20 | |
| 53 | GPT-5.4 MiniOpenAI | 83.1% | 1m 03s | 35.8 | $4.50 | |
| 54 | Qwen3.5-FlashQwen | 82.8% | 1m 20s | 41.1 | $0.26 | |
| 55 | MiMo-V2.5-ProXiaomi | 82.6% | 4m 11s | 31.6 | $0.9152 | |
| 56 | MiniMax M2.5MiniMax | 82.1% | 2m 52s | 37.2 | $1.20 | |
| 57 | Claude Sonnet 4.5Anthropic | 81.6% | 1m 57s | 39.5 | $15.00 | |
| 58 | MiMo-V2.5Xiaomi | 81.6% | 1m 29s | 50.0 | $0.336 | |
| 59 | Gemini 3.1 Flash Lite PreviewGoogle | 81.1% | 16 s | 22.7 | $1.50 | |
| 60 | DeepSeek V3.2DeepSeek | 80.3% | 4m 20s | 30.0 | $0.88 | |
| 61 | GPT-5 MiniOpenAI | 80.3% | 1m 07s | 34.2 | $2.00 | |
| 62 | GLM 4.7Z.ai | 80.0% | 3m 31s | 55.1 | $2.20 | |
| 63 | MiniMax M2.1MiniMax | 78.5% | 2m 15s | Not benchmarked | $1.20 | |
| 64 | Kimi K2 ThinkingMoonshot AI | 78.5% | 9m 46s | Not benchmarked | $2.50 | |
| 65 | gpt-oss-120bOpenAI | 78.5% | 56 s | Not benchmarked | $0.50 | |
| 66 | GPT-5.4 NanoOpenAI | 77.5% | 30 s | 44.8 | $1.25 | |
| 67 | Claude Opus 4.1Anthropic | 76.3% | 36 s | Not benchmarked | $75.00 | |
| 68 | o3 MiniOpenAI | 75.5% | 1m 40s | Not benchmarked | $4.40 | |
| 69 | o4 MiniOpenAI | 74.5% | 30 s | Not benchmarked | $4.40 | |
| 70 | GLM 4.6Z.ai | 74.5% | 7m 55s | 8.8 | $2.20 | |
| 71 | Grok 3RetiredxAI | 74.2% | 30 s | Not benchmarked | $15.00 | |
| 72 | o1OpenAI | 73.2% | 40 s | Not benchmarked | $60.00 | |
| 73 | Claude Haiku 4.5Anthropic | 72.2% | 1m 23s | 35.6 | $5.00 | |
| 74 | GLM 4.5Z.ai | 72.2% | 11m 44s | Not benchmarked | $2.20 | |
| 75 | Kimi K2 0711Moonshot AI | 71.5% | 41 s | Not benchmarked | $2.30 | |
| 76 | Gemini 2.5 Flash LiteGoogle | 70.2% | 21 s | Not benchmarked | $0.40 | |
| 77 | Llama 4 MaverickMeta Llama | 69.4% | 10 s | Not benchmarked | $1.00 | |
| 78 | gpt-oss-20bOpenAI | 68.9% | 2m 25s | Not benchmarked | $0.15 | |
| 79 | Mistral Large 3 2512Mistral AI | 68.4% | 21 s | Not benchmarked | $1.50 | |
| 80 | GPT-4.1 MiniOpenAI | 67.9% | 9.3 s | Not benchmarked | $1.60 | |
| 81 | GPT-4.1OpenAI | 65.4% | 23 s | Not benchmarked | $8.00 | |
| 82 | Gemini 2.0 FlashRetiredGoogle | 65.2% | 5.8 s | Not benchmarked | $0.40 | |
| 83 | GPT-5 NanoOpenAI | 63.4% | 50 s | Not benchmarked | $0.40 | |
| 84 | DeepSeek V3 0324DeepSeek | 61.6% | 23 s | Not benchmarked | $1.07 | |
| 85 | Gemini 1.5 ProRetiredGoogle | 58.3% | 7.7 s | Not benchmarked | $5.00 | |
| 86 | GPT-4o (2024-11-20)OpenAI | 53.8% | 15 s | Not benchmarked | $10.00 | |
| 87 | GPT-4.1 NanoOpenAI | 50.8% | 4.2 s | Not benchmarked | $0.40 | |
| 88 | GPT-4o (2024-05-13)OpenAI | 50.2% | 10 s | Not benchmarked | $15.00 | |
| 89 | Command ACohere | 48.5% | 11 s | Not benchmarked | $10.00 | |
| 90 | Mistral LargeMistral AI | 47.7% | 16 s | Not benchmarked | $6.00 | |
| 91 | Llama 4 ScoutMeta Llama | 47.0% | 12 s | Not benchmarked | $0.50 | |
| 92 | Gemini 1.5 FlashRetiredGoogle | 46.0% | 3.4 s | Not benchmarked | $0.30 | |
| 93 | GPT-4o-miniOpenAI | 44.2% | 15 s | Not benchmarked | $0.60 | |
| 94 | Mistral Medium 3.5Mistral AI | 34.9% | 4m 16s | 25.5 | $7.50 | |
| 95 | GPT-3.5 TurboOpenAI | 30.6% | 2.6 s | Not benchmarked | $1.50 |
Only models in our catalog are listed, and each appears once with its best published run. 95 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out. Every other column is a figure the board publishes for the same run.