benchgap
Calibration

Toolathlon-Verified → JobBench

JobBench is estimated from Toolathlon-Verified with a Hill curve fitted on 7 models measured on both: y = 0.0000 + (1.0000 − 0.0000)·x^6.00 / (0.70094^6.00 + x^6.00), R² = 0.49, cross-validated error 3.4 pp. It is used for 14 estimates.

Estimated modelToolathlon-VerifiedJobBenchSource
Claude Fable 5.177.8%65.2%estimated ± 3.4 pp, low confidence
Claude Opus 580.6%69.8%estimated ± 3.4 pp, low confidence
Claude Opus 5.577.8%65.2%estimated ± 3.4 pp, low confidence
Claude Sonnet 5.577.8%65.2%estimated ± 3.4 pp, low confidence
DeepSeek V4 Flash 073170.3%50.4%estimated ± 3.4 pp, low confidence
DeepSeek V4 Pro 081374.1%58.3%estimated ± 3.4 pp, low confidence
dots3-note Preview55.6%19.9%estimated ± 3.4 pp, low confidence
GLM-5.373.0%56.1%estimated ± 3.4 pp, low confidence
GLM-5.3-Flash78.4%66.2%estimated ± 3.4 pp, low confidence
Inkling-Small54.4%17.9%estimated ± 3.4 pp, low confidence
Laguna S 2.149.7%11.3%estimated ± 3.4 pp, low confidence
Ornith-1.5-35B-A3B48.7%10.1%estimated ± 3.4 pp, low confidence
Ornith-1.5-397B71.2%52.3%estimated ± 3.4 pp, low confidence
Ornith-1.5-9B41.2%4.0%estimated ± 3.4 pp, low confidence