Skip to content

Benchmark

GPQA Diamond

Science benchmarks measure how well models answer science questions. Scores are percentages. Higher is better.

Models and systems measured
282
Scores recorded
1623
Sources
4
IQ weight
1.72

Score distribution

79.3%
Mean
84.5%
Median
15.7%
Std dev
56.8%
10th percentile
92.2%
90th percentile
0.168%
Min
0.168%96.8%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.72 for chat models, measured over 272 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
1501GLM 4.7 FlashOpenRouter harness, z-ai/glm-4.7-flash-20260119, Venice, public 90-day aggregate· OpenRouter benchmark exports49.8%
1502Llama 3.3 70B InstructOpenRouter harness, meta-llama/llama-3.3-70b-instruct, Groq, public 90-day aggregate· OpenRouter benchmark exports49.5%
1503Llama 3.3 70B InstructOpenRouter harness, meta-llama/llama-3.3-70b-instruct, Cloudflare, public 90-day aggregate· OpenRouter benchmark exports49.3%
1504Qwen3 14BOpenRouter harness, qwen/qwen3-14b-04-28, Alibaba, public 90-day aggregate· OpenRouter benchmark exports49.2%
1505Qwen2.5 72B Instruct· Epoch AI Benchmarking Hub data49.1%
1506GPT-4.1 nanoOpenRouter harness, openai/gpt-4.1-nano-2025-04-14, auto-routing, public 90-day aggregate· OpenRouter benchmark exports49.1%
1507Mistral Small 3.2· Epoch AI Benchmarking Hub data49.1%
1508Mistral Large2407· Epoch AI Benchmarking Hub data49%
1509GPT-4.1 nano· Epoch AI Benchmarking Hub data48.9%
1510Llama 3.3 70B InstructOpenRouter harness, meta-llama/llama-3.3-70b-instruct, CoreWeave, public 90-day aggregate· OpenRouter benchmark exports48.7%
1511GLM 4.7 FlashOpenRouter harness, z-ai/glm-4.7-flash-20260119, DeepInfra, public 90-day aggregate· OpenRouter benchmark exports48.7%
1512GPT-5 nanominimal· Epoch AI Benchmarking Hub data48.5%
1513Llama 3.3 70B InstructOpenRouter harness, meta-llama/llama-3.3-70b-instruct, AkashML, public 90-day aggregate· OpenRouter benchmark exports48.3%
1514Llama 3.3 70B InstructOpenRouter harness, meta-llama/llama-3.3-70b-instruct, auto-routing, public 90-day aggregate· OpenRouter benchmark exports48.2%
1515Llama 3.3 70B InstructOpenRouter harness, meta-llama/llama-3.3-70b-instruct, SambaNova, public 90-day aggregate· OpenRouter benchmark exports48.1%
1516Qwen Plus· Epoch AI Benchmarking Hub data48.1%
1517GPT 4o· Epoch AI Benchmarking Hub data47.9%
1518Llama 3.3 70B InstructOpenRouter harness, meta-llama/llama-3.3-70b-instruct, Parasail, public 90-day aggregate· OpenRouter benchmark exports47.9%
1519Gemma 3 27B IT· Epoch AI Benchmarking Hub data47.7%
1520Llama 3.3 70B InstructOpenRouter harness, meta-llama/llama-3.3-70b-instruct, Together, public 90-day aggregate· OpenRouter benchmark exports47.7%
1521Qwen2.5 72B InstructOpenRouter harness, qwen/qwen-2.5-72b-instruct, DeepInfra, public 90-day aggregate· OpenRouter benchmark exports47.7%
1522Magistral Small 2509· Epoch AI Benchmarking Hub data47.6%
1523Mistral Small2503· Epoch AI Benchmarking Hub data47.5%
1524Qwen3 8Bnone· Epoch AI Benchmarking Hub data47.5%
1525Llama 3.3 70B Instruct· Epoch AI Benchmarking Hub data47.4%
1526gemini-1.5-flash-002· Epoch AI Benchmarking Hub data47.3%
1527Mistral-Small-3· Epoch AI Benchmarking Hub data47.3%
1528Llama 3.3 70B InstructOpenRouter harness, meta-llama/llama-3.3-70b-instruct, Novita, public 90-day aggregate· OpenRouter benchmark exports47.2%
1529Claude 3 Opus· Epoch AI Benchmarking Hub data47.2%
1530Llama 3.3 70B InstructOpenRouter harness, meta-llama/llama-3.3-70b-instruct, PrimeIntellect, public 90-day aggregate· OpenRouter benchmark exports47%
1531Llama 3.3 70B InstructOpenRouter harness, meta-llama/llama-3.3-70b-instruct, Crusoe, public 90-day aggregate· OpenRouter benchmark exports46.9%
1532GPT-4 Turbo· Epoch AI Benchmarking Hub data46.6%
1533Llama 3.3 70B InstructOpenRouter harness, meta-llama/llama-3.3-70b-instruct, DeepInfra, public 90-day aggregate· OpenRouter benchmark exports46.3%
1534Qwen2.5 72B InstructOpenRouter harness, qwen/qwen-2.5-72b-instruct, Novita, public 90-day aggregate· OpenRouter benchmark exports46.1%
1535Qwen2.5 32B Instruct· Epoch AI Benchmarking Hub data46.1%
1536Qwen2.5 72B InstructOpenRouter harness, qwen/qwen-2.5-72b-instruct, auto-routing, public 90-day aggregate· OpenRouter benchmark exports46%
1537DeepSeek V3OpenRouter harness, deepseek/deepseek-chat-v3-0324, auto-routing, public 90-day aggregate· OpenRouter benchmark exports46%
1538GPT OSS 20bhigh· Epoch AI Benchmarking Hub data46%
1539GPT OSS 120BOpenRouter harness, openai/gpt-oss-120b, DigitalOcean, public 90-day aggregate· OpenRouter benchmark exports45.9%
1540gemini-1.5-pro-001· Epoch AI Benchmarking Hub data45.9%
1541Qwen2.5 VL 72B InstructOpenRouter harness, qwen/qwen2.5-vl-72b-instruct, Parasail, public 90-day aggregate· OpenRouter benchmark exports45.7%
1542Llama 3.3 70B InstructOpenRouter harness, meta-llama/llama-3.3-70b-instruct, Nebius, public 90-day aggregate· OpenRouter benchmark exports45.6%
1543Mistral Small· Epoch AI Benchmarking Hub data45.3%
1544GLM 4.7 Flashnone· Epoch AI Benchmarking Hub data45.1%
1545DeepSeek R1 Distill Qwen 14B· Epoch AI Benchmarking Hub data44.7%
1546Qwen2.5 VL 72B InstructOpenRouter harness, qwen/qwen2.5-vl-72b-instruct, Nebius, public 90-day aggregate· OpenRouter benchmark exports44.3%
1547Llama 3.1 70B Instruct· Epoch AI Benchmarking Hub data44.2%
1548GPT-4o miniOpenRouter harness, openai/gpt-4o-mini, auto-routing, public 90-day aggregate· OpenRouter benchmark exports43.6%
1549Qwen3 4Bnone· Epoch AI Benchmarking Hub data43.6%
1550WizardLM 2 8x22B· Epoch AI Benchmarking Hub data43.4%

Showing 1,501–1,550 of 1,621 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
MiMo V2 Flash Baseofficial Base table · 5-shot· Reported by Xiaomi MiMo· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.55.1%
Qwen3.8-2.4T-A95Blifecycle-page core-benchmark claim· Reported by Alibaba Cloud Model Studio· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.92.6%

Tokens the model generated per task, as measured by the source. Comparable within this benchmark only.

Data credits: Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · 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. · OpenRouter public benchmark leaderboard and search-lane exports, CC BY 4.0. Fractional source scores converted to percentages; immutable dated snapshots. (licence) · OpenEvals/leaderboard-data, MIT (licence)