Skip to content

Benchmark

HELM Capabilities: GPQA - COT correct

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

Models measured
57
Scores recorded
59
Sources
1
IQ weight
1.18

Score distribution

54.4%
Mean
55.6%
Median
15.1%
Std dev
33.4%
10th percentile
72.6%
90th percentile
24.7%
Min
24.7%80.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.18 for chat models, measured over 57 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Gemini 3 Pro Preview· Stanford HELM specialized suites data docs80.3%
2GPT 5· Stanford HELM specialized suites data docs79.1%
3GPT-5 mini· Stanford HELM specialized suites data docs75.6%
4OpenAI o3· Stanford HELM specialized suites data docs75.3%
5Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs74.9%
6o4 Mini· Stanford HELM specialized suites data docs73.5%
7Grok 4· Stanford HELM specialized suites data docs72.6%
8Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs72.6%
9Claude Opus 4thinking, 10k· Stanford HELM specialized suites data docs70.9%
10Claude Sonnet 4thinking, 10k· Stanford HELM specialized suites data docs70.6%
11Claude Sonnet 4.5· Stanford HELM specialized suites data docs68.6%
12GPT OSS 120B· Stanford HELM specialized suites data docs68.4%
13GPT-5 nano· Stanford HELM specialized suites data docs67.9%
14Grok 3 Mini· Stanford HELM specialized suites data docs67.5%
15Claude Opus 4setting not named by the source· Stanford HELM specialized suites data docs66.6%
16DeepSeek R1· Stanford HELM specialized suites data docs66.6%
17Palmyra X5· Stanford HELM specialized suites data docs66.1%
18GPT 4.1· Stanford HELM specialized suites data docs65.9%
19Kimi K2 Instruct· Stanford HELM specialized suites data docs65.2%
20Grok 3· Stanford HELM specialized suites data docs65%
21Llama 4 Maverick· Stanford HELM specialized suites data docs65%
22Claude Sonnet 4setting not named by the source· Stanford HELM specialized suites data docs64.3%
23Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs63%
24GPT-4.1 mini· Stanford HELM specialized suites data docs61.4%
25Claude Sonnet 3.7· Stanford HELM specialized suites data docs60.8%
26Claude Haiku 4.5· Stanford HELM specialized suites data docs60.5%
27GLM 4.5 Air· Stanford HELM specialized suites data docs59.4%
28GPT OSS 20b· Stanford HELM specialized suites data docs59.4%
29Claude 3.5 Sonnet· Stanford HELM specialized suites data docs56.5%
30Gemini 2.0 Flash· Stanford HELM specialized suites data docs55.6%
31DeepSeek V3· Stanford HELM specialized suites data docs53.8%
32gemini-1.5-pro-002· Stanford HELM specialized suites data docs53.4%
33Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs52.2%
34GPT 4o· Stanford HELM specialized suites data docs52%
35Nova Premier 1.0· Stanford HELM specialized suites data docs51.8%
36GPT-4.1 nano· Stanford HELM specialized suites data docs50.7%
37Llama 4 Scout· Stanford HELM specialized suites data docs50.7%
38Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs50%
39Nova Prov1· Stanford HELM specialized suites data docs44.6%
40GPT 5.1· Stanford HELM specialized suites data docs44.2%
41gemini-1.5-flash-002· Stanford HELM specialized suites data docs43.7%
42Mistral Large2411· Stanford HELM specialized suites data docs43.5%
43Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs42.6%
44Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs42.6%
45Nova Litev1· Stanford HELM specialized suites data docs39.7%
46Mistral Small· Stanford HELM specialized suites data docs39.2%
47gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs39%
48Granite 4.0 H Small· Stanford HELM specialized suites data docs38.3%
49Nova Microv1· Stanford HELM specialized suites data docs38.3%
50GPT-4o mini· Stanford HELM specialized suites data docs36.8%

Showing 1–50 of 59 results.

Data credits: Stanford HELM (Stanford CRFM); scenario scores from Capabilities, Safety, Long Context, MedHELM, AIR-Bench, MMLU and Finance