benchgap
Calibration

GDPval-AA → MCP Atlas

MCP Atlas is estimated from GDPval-AA with a linear curve fitted on 28 models measured on both: y = 0.6059·x + 0.5017, R² = 0.36, cross-validated error 11.1 pp. It is used for 27 estimates.

Estimated modelGDPval-AAMCP AtlasSource
A.X K223.2%64.2%estimated ± 11.1 pp, low confidence
DeepSeek V3 03240.0%50.2%estimated ± 11.1 pp, low confidence
Gemma 4 12B0.0%50.2%estimated ± 11.1 pp, low confidence
Gemma 4 26B A4B3.4%52.2%estimated ± 11.1 pp, low confidence
Gemma 4 E2B0.0%50.2%estimated ± 11.1 pp, low confidence
Gemma 4 E4B0.0%50.2%estimated ± 11.1 pp, low confidence
GPT-4.1 mini0.0%50.2%estimated ± 11.1 pp, low confidence
GPT-4.1 nano0.0%50.2%estimated ± 11.1 pp, low confidence
GPT-4o0.0%50.2%estimated ± 11.1 pp, low confidence
GPT-4o mini0.0%50.2%estimated ± 11.1 pp, low confidence
GPT-5.116.5%60.2%estimated ± 11.1 pp, low confidence
Granite 4.2 30B3.9%52.5%estimated ± 11.1 pp, low confidence
Granite 4.2 3B0.0%50.2%estimated ± 11.1 pp, low confidence
K-Exaone0.0%50.2%estimated ± 11.1 pp, low confidence
LFM2.5-2.6B0.0%50.2%estimated ± 11.1 pp, low confidence
Ling 2.6 Flash0.0%50.2%estimated ± 11.1 pp, low confidence
Ling 3.0 Flash VL33.2%70.3%estimated ± 11.1 pp, low confidence
Ling 3.0 Tiny3.2%52.1%estimated ± 11.1 pp, low confidence
Ling 3.1 Flash56.1%84.2%estimated ± 11.1 pp, low confidence
Mercury 2.50.0%50.2%estimated ± 11.1 pp, low confidence
MiMo-V2-Flash6.6%54.2%estimated ± 11.1 pp, low confidence
MiniCPM5-2B10.5%56.5%estimated ± 11.1 pp, low confidence
Muse Spark25.1%65.4%estimated ± 11.1 pp, low confidence
Nemotron 3 Nano Omni 30B A3B0.0%50.2%estimated ± 11.1 pp, low confidence
North Mini Code0.0%50.2%estimated ± 11.1 pp, low confidence
Solar Pro 30.0%50.2%estimated ± 11.1 pp, low confidence
Ultravox v0.6 Llama 3.3 70B0.0%50.2%estimated ± 11.1 pp, low confidence