Benchmark
GPQA
Science benchmarks measure how well models answer science questions. Scores are percentages. Higher is better.
- Models and systems measured
- 364
- Scores recorded
- 378
- Sources
- 3
- IQ weight
- 0.75
Score distribution
- 45.3%
- Mean
- 47.5%
- Median
- 34.4%
- Std dev
- 2.9%
- 10th percentile
- 89%
- 90th percentile
- 0%
- Min
This benchmark has a weight of 0.75 for chat models, measured over 359 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model or system | Score |
|---|---|---|
| 51 | Grok 4· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 87.5% |
| 52 | DeepSeek V4 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 87.4% |
| 53 | GPT 5high· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 87.3% |
| 54 | Claude Opus 4.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 87% |
| 55 | Nemotron 3 Ultra· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 87% |
| 56 | Nemotron 3 UltraNeMo Evaluator SDK · no tools· Reported by NVIDIA· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 87% |
| 57 | Gemini 3.1 Flash Lite· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 86.9% |
| 58 | Qwen3.5 122B-A10B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 86.6% |
| 59 | Gemini 2.5 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 86.4% |
| 60 | GLM 5.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 86.2% |
| 61 | Qwen3.6 35B A3B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 86% |
| 62 | GLM 4.7· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 85.7% |
| 63 | GPT 5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 85.7% |
| 64 | Grok 4 Fast· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 85.7% |
| 65 | GPT-5.5 Instant· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 85.6% |
| 66 | Qwen3.5 27B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 85.5% |
| 67 | Seed 2.0 Lite· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 85.1% |
| 68 | ERNIE 5.0· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 85% |
| 69 | Claude Sonnet 3.7· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 84.8% |
| 70 | Grok 3· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 84.6% |
| 71 | Kimi K2 Thinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 84.5% |
| 72 | Gemma 4 31B IT· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 84.3% |
| 73 | Mai Thinking 1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 84.2% |
| 74 | Qwen3.5 35B A3B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 84.2% |
| 75 | ChatGPT-4o Latest· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 84% |
| 76 | Grok 3 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 84% |
| 77 | Qwen3.5 Omni PlusQwen3.5-Omni release · no-thinking setting· Reported by Qwen, Qwen Team· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 83.9% |
| 78 | MiMo V2 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 83.7% |
| 79 | Muse Glimmer 30B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 83.5% |
| 80 | Claude Sonnet 4.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 83.4% |
| 81 | OpenAI o3· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 83.3% |
| 82 | Gemini 2.5 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 82.8% |
| 83 | GPT 5.4 Nano· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 82.8% |
| 84 | Nemotron 3 Super 120B A12B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 82.7% |
| 85 | Nemotron 3 Super 120B A12BNeMo Evaluator / mostly NeMo Skills · with tools· Reported by NVIDIA· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 82.7% |
| 86 | DeepSeek Reasoner· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 82.4% |
| 87 | DeepSeek V3.2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 82.4% |
| 88 | Gemma 4 26B A4B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 82.3% |
| 89 | GPT-5 mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 82.3% |
| 90 | Qwen3.5 9B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 81.7% |
| 91 | LongCat-Flash-Thinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 81.5% |
| 92 | o4 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 81.4% |
| 93 | Qwen 3 235B A22B Thinking 2507· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 81.1% |
| 94 | DeepSeek R1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 81% |
| 95 | GLM 4.6· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 81% |
| 96 | MiniMax M2.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 81% |
| 97 | Claude Opus 4.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 80.9% |
| 98 | GPT OSS 120Bhigh· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 80.9% |
| 99 | GPT OSS 120B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 80.1% |
| 100 | DeepSeek V3.2 Exp· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 79.9% |
Showing 51–100 of 374 results.
Not ranked
The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.
| Model or system | Score |
|---|---|
| Qwen3.5-397B-A17B-BaseQwen3.5 release/model card · base-model table· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 54.6% |
| Qwen3.7-Plus-BaseQwen3.8-Flash-Next release/model card · base-model table· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 51.5% |
| Qwen3.8-27B-BaseQwen3.8-Flash-Next release/model card · base-model table· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 45% |
| Qwen3.8-Flash-Next-BaseQwen3.8-Flash-Next release/model card · base-model table· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 51.4% |
Data credits: Data by LLM Stats · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.