benchgap
Calibration

SkillsBench → WideResearch

WideResearch is estimated from SkillsBench with a Hill curve fitted on 6 models measured on both: y = 0.3703 + (0.8383 − 0.3703)·x^6.00 / (0.36277^6.00 + x^6.00), R² = 0.93, cross-validated error 1.3 pp. It is used for 1 estimate.

Estimated modelSkillsBenchWideResearchSource
Muse Glimmer 30B44.3%73.0%estimated ± 1.3 pp, low confidence

Fitted on

ModelSkillsBenchWideResearch
Atria Dawn Preview66.4%81.9%
dots3-note Preview52.8%78.9%
Hy4 preview62.9%83.9%
Ling 3.0 Flash44.8%73.6%
Ling 3.1 Flash68.7%83.3%
Qwen3.8 Max70.2%81.9%