Knowledge benchmark
GPQA Diamond (Vals) leaderboard
GPQA Diamond, Vals AI run. Every model the catalog carries a published GPQA Diamond (Vals) value for, ranked by that value.
CategoryKnowledge
MeasureAccuracy
TasksGraduate-level science questions
DifficultyExpert reasoning
Published byVals AI GPQA Diamond, Vals AI run leaderboard
Vals AI’s independent run of the GPQA Diamond graduate-level science questions.
GPQA Diamond (Vals) ranking
50 models with a published GPQA Diamond (Vals) value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.
| Rank | Model | Provider | Accuracy |
|---|---|---|---|
| 1 |  Gemini 3.1 Pro | 95.5 | |
| 2 |  GPT-5.6 Sol | OpenAI | 95.2 |
| 3 |  Grok 4.6 | xAI | 94.7 |
| 4 | Gemini 3.8 Flash | 94.4 | |
| 5 | Gemini 3.7 Flash | 93.9 | |
| 6 |  Qwen3.8 Max | Alibaba | 93.7 |
| 7 |  Claude Fable 5.1 | Anthropic | 93.4 |
| 7 | Claude Opus 5 | Anthropic | 93.4 |
| 7 | Gemini 3.6 Flash | 93.4 | |
| 10 | Claude Fable 5 | Anthropic | 93.2 |
| 10 | GPT-5.5 | OpenAI | 93.2 |
| 12 | Grok 4.5 | xAI | 92.9 |
| 12 |  Kimi K3 | Moonshot AI | 92.9 |
| 14 | Gemini 3.5 Flash | 92.7 | |
| 14 |  MiniMax M3 | MiniMax | 92.7 |
| 16 | Claude Opus 4.8 | Anthropic | 92.4 |
| 16 |  DeepSeek V4 Pro 0813 | DeepSeek | 92.4 |
| 18 | GPT-5.6 Luna | OpenAI | 91.7 |
| 19 | Grok 4.3 | xAI | 91.4 |
| 20 |  Muse Spark 1.1 | Meta | 91.2 |
| 21 | GPT-5.6 Terra | OpenAI | 90.9 |
| 22 | Claude Opus 4.7 | Anthropic | 90.2 |
| 22 | Qwen3.7 Max | Alibaba | 90.2 |
| 24 | DeepSeek V4 Flash 0731 | DeepSeek | 89.9 |
| 25 | Kimi K2.6 | Moonshot AI | 89.1 |
| 26 | Claude Sonnet 5 | Anthropic | 88.9 |
| 26 | Qwen3.8-27B | Alibaba | 88.9 |
| 28 | Grok 4.20 | xAI | 88.6 |
| 29 |  GLM-5.3 | Z.AI | 88.1 |
| 30 | Gemini 3 Flash | 87.9 | |
| 31 | Qwen3.6 Plus | Alibaba | 87.4 |
| 32 | TMInkling | Thinking Machines Lab | 87.1 |
| 33 | MiniMax M2.7 | MiniMax | 86.6 |
| 34 | GLM-5.3-Flash | Z.AI | 86.4 |
| 35 |  Nemotron 3 Ultra | NVIDIA | 86.1 |
| 36 | Claude Sonnet 4.6 | Anthropic | 85.6 |
| 36 | GLM-5.2 | Z.AI | 85.6 |
| 38 |  Ling 3.0 Flash | InclusionAI | 84.8 |
| 39 | GLM-5.1 | Z.AI | 84.5 |
| 40 | Gemini 3.5 Flash-Lite | 83.8 | |
| 41 | TMInkling-Small | Thinking Machines Lab | 83.6 |
| 42 | GPT-5.4 mini | OpenAI | 83.1 |
| 43 |  MiMo-V2.5-Pro | Xiaomi | 82.6 |
| 44 | MiMo-V2.5 | Xiaomi | 81.6 |
| 45 | Gemini 3.1 Flash-Lite | 81.1 | |
| 46 | GPT-5.4 nano | OpenAI | 77.5 |
| 47 | Claude Haiku 4.5 | Anthropic | 72.2 |
| 48 |  Laguna XS.2 | Poolside | 55.1 |
| 49 |  Mistral Medium 3.5 128B | Mistral | 34.8 |
| 50 | Laguna M.1 | Poolside | 27.0 |
Evidence key: Observed
Rows are ordered by the value Vals AI GPQA Diamond, Vals AI run leaderboard published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.