benchgap
Calibration

Toolathlon → WideResearch

WideResearch is estimated from Toolathlon with a linear curve fitted on 7 models measured on both: y = 0.6351·x + 0.4879, R² = 0.65, cross-validated error 5.6 pp. It is used for 1 estimate.

Estimated modelToolathlonWideResearchSource
MiniMax M2.746.3%78.2%estimated ± 5.6 pp, low confidence

Fitted on

ModelToolathlonWideResearch
Claude Opus 4.543.5%76.4%
GLM-538.0%69.8%
Kimi K2.650.0%80.8%
Kimi K2.527.8%72.7%
Qwen3.5 397B36.3%74.0%
Qwen3.6-35B-A3B26.9%60.1%
Qwen3.6 Plus39.8%74.3%