Benchmark
Vals GPQA Diamond
Science benchmarks measure how well models answer science questions. Scores are percentages. Higher is better.
- Models measured
- 121
- Scores recorded
- 126
- Sources
- 1
- IQ weight
- 2.35
Score distribution
- 76%
- Mean
- 81.6%
- Median
- 16.7%
- Std dev
- 50.8%
- 10th percentile
- 92.7%
- 90th percentile
- 27%
- Min
This benchmark has a weight of 2.35 for chat models, measured over 121 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 101 | GPT-5 nano· Vals AI | 63.4% |
| 102 | Magistral Medium· Vals AI | 62.4% |
| 103 | DeepSeek V3· Vals AI | 61.6% |
| 104 | Claude 3.5 Sonnet· Vals AI | 59.3% |
| 105 | MiMo V2 Flash· Vals AI | 59.3% |
| 106 | gemini-1.5-pro-002· Vals AI | 58.3% |
| 107 | Magistral Small 2509· Vals AI | 58.3% |
| 108 | gemini-2.5-flash-preview-04-17· Vals AI | 57.6% |
| 109 | Laguna XS.2· Vals AI | 55.1% |
| 110 | GPT 4o· Vals AI | 53.8% |
| 111 | GPT-4.1 nano· Vals AI | 50.8% |
| 112 | Grok 2· Vals AI | 50.8% |
| 113 | Mistral Small· Vals AI | 50.8% |
| 114 | Llama 3.3 70B Instruct Turbo· Vals AI | 50% |
| 115 | Command A 03 2025· Vals AI | 48.5% |
| 116 | Mistral Large2411· Vals AI | 47.7% |
| 117 | Llama 4 Scout· Vals AI | 47% |
| 118 | gemini-1.5-flash-002· Vals AI | 46% |
| 119 | GPT-4o mini· Vals AI | 44.2% |
| 120 | Claude Haiku 3.5· Vals AI | 37.9% |
| 121 | jamba-1.6-large· Vals AI | 36.1% |
| 122 | Mistral Medium 3.5· Vals AI | 34.8% |
| 123 | jamba-1.6-mini· Vals AI | 32.1% |
| 124 | Command R Plus· Vals AI | 31.1% |
| 125 | GPT-3.5 Turbo· Vals AI | 30.6% |
| 126 | Laguna M.1· Vals AI | 27% |
Showing 101–126 of 126 results.