benchgap
Calibration

HLE → Artificial Analysis Intelligence Index

Artificial Analysis Intelligence Index is estimated from HLE with a linear curve fitted on 52 models measured on both: y = 0.6182·x + 0.0357, R² = 0.47, cross-validated error 10.7 pp. It is used for 3 estimates.

Estimated modelHLEArtificial Analysis Intelligence IndexSource
Hy459.1%40.1%estimated ± 10.7 pp, low confidence
Hy4 flash55.2%37.7%estimated ± 10.7 pp, low confidence
Agents-A147.6%33.0%estimated ± 10.7 pp, low confidence

Fitted on

ModelHLEArtificial Analysis Intelligence Index
A.X K227.8%9.0%
Apodex 1.156.1%13.7%
Claude Fable 5.165.0%49.2%
Claude Fable 5.274.2%64.6%
Claude Opus 4.530.8%23.7%
Claude Opus 4.653.0%26.4%
Claude Opus 4.754.7%40.7%
Claude Opus 4.857.9%41.8%
Claude Opus 564.7%50.8%
Claude Sonnet 4.649.0%24.7%
Claude Sonnet 557.4%38.2%
DeepSeek V4.1 Flash 101742.6%39.5%
DeepSeek V4 Flash 073134.8%34.3%
DeepSeek V4 Pro42.7%53.2%
Gemini 2.5 Pro18.8%16.1%
Gemini 3.5 Flash40.2%50.2%
Gemma 4 31B26.5%4.4%
GLM-4.724.8%22.2%
GLM-530.5%27.9%
GLM-5.131.0%26.1%
GLM-5.240.5%33.7%
GPT-5.452.1%39.0%
GPT-5.4 mini41.5%24.1%
GPT-5.4 nano37.7%20.7%
GPT-5.552.2%38.4%
GPT-OSS 120B14.9%3.8%
Grok 4.335.0%37.6%
Hy3 Preview25.5%41.2%
Inkling46.0%12.5%
Inkling 2 Small53.2%30.6%
Inkling-Small47.8%12.9%
K-EXAONE 2.018.3%7.6%
Kimi K2.634.7%27.0%
Kimi K3.1 Pro70.0%46.5%
Kimi K24.7%12.7%
Kimi K2.530.1%23.5%
Kimi K356.0%43.6%
Ling 3.0 Flash22.7%20.1%
MiMo-V2.5-Pro48.0%26.0%
MiniMax M2.728.0%22.8%
Muse Spark50.4%31.3%
Muse Spark 1.162.1%33.7%
Nemotron 3.5 Super VL 120B A12B29.2%13.6%
Nemotron 3.5 Super VL 120B A12B NVFP429.0%13.6%
Nemotron 3 Ultra26.7%11.8%
Qwen3.6-27B24.0%21.4%
Qwen3.6 Plus28.8%27.0%
Qwen3.7 Max41.4%29.5%
Qwen3.8-27B30.8%20.1%
Qwen3.8-Flash-Next35.9%28.6%
Qwen4 Max51.7%42.9%
Step 5 Preview46.5%28.5%