benchgap
Calibration

Artificial Analysis Intelligence Index → HLE

HLE is estimated from Artificial Analysis Intelligence Index with a linear curve fitted on 52 models measured on both: y = 0.7651·x + 0.1853, R² = 0.47, cross-validated error 12.0 pp. It is used for 1 estimate.

Estimated modelArtificial Analysis Intelligence IndexHLESource
Claude 4.1 Opus18.6%32.7%estimated ± 12.0 pp, low confidence

Fitted on

ModelArtificial Analysis Intelligence IndexHLE
A.X K29.0%27.8%
Apodex 1.113.7%56.1%
Claude Fable 5.149.2%65.0%
Claude Fable 5.264.6%74.2%
Claude Opus 4.523.7%30.8%
Claude Opus 4.626.4%53.0%
Claude Opus 4.740.7%54.7%
Claude Opus 4.841.8%57.9%
Claude Opus 550.8%64.7%
Claude Sonnet 4.624.7%49.0%
Claude Sonnet 538.2%57.4%
DeepSeek V4.1 Flash 101739.5%42.6%
DeepSeek V4 Flash 073134.3%34.8%
DeepSeek V4 Pro53.2%42.7%
Gemini 2.5 Pro16.1%18.8%
Gemini 3.5 Flash50.2%40.2%
Gemma 4 31B4.4%26.5%
GLM-4.722.2%24.8%
GLM-527.9%30.5%
GLM-5.126.1%31.0%
GLM-5.233.7%40.5%
GPT-5.439.0%52.1%
GPT-5.4 mini24.1%41.5%
GPT-5.4 nano20.7%37.7%
GPT-5.538.4%52.2%
GPT-OSS 120B3.8%14.9%
Grok 4.337.6%35.0%
Hy3 Preview41.2%25.5%
Inkling12.5%46.0%
Inkling 2 Small30.6%53.2%
Inkling-Small12.9%47.8%
K-EXAONE 2.07.6%18.3%
Kimi K2.627.0%34.7%
Kimi K3.1 Pro46.5%70.0%
Kimi K212.7%4.7%
Kimi K2.523.5%30.1%
Kimi K343.6%56.0%
Ling 3.0 Flash20.1%22.7%
MiMo-V2.5-Pro26.0%48.0%
MiniMax M2.722.8%28.0%
Muse Spark31.3%50.4%
Muse Spark 1.133.7%62.1%
Nemotron 3.5 Super VL 120B A12B13.6%29.2%
Nemotron 3.5 Super VL 120B A12B NVFP413.6%29.0%
Nemotron 3 Ultra11.8%26.7%
Qwen3.6-27B21.4%24.0%
Qwen3.6 Plus27.0%28.8%
Qwen3.7 Max29.5%41.4%
Qwen3.8-27B20.1%30.8%
Qwen3.8-Flash-Next28.6%35.9%
Qwen4 Max42.9%51.7%
Step 5 Preview28.5%46.5%