Calibration

Vals SWE-bench → FrontierCode 1.1 Extended

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap
License: https://creativecommons.org/licenses/by-nc/4.0/

FrontierCode 1.1 Extended is estimated from Vals SWE-bench with a Hill curve fitted on 5 models measured on both: y = 0.0000 + (1.2000 − 0.0000)·x^6.00 / (0.95815^6.00 + x^6.00), R² = 0.69, cross-validated error 2.4 pp. It is used for 3 estimates.

Estimated model	Vals SWE-bench	FrontierCode 1.1 Extended	Source
DeepSeek V4 Pro 0813	96.4%	61.1%	estimated ± 2.4 pp, medium confidence
GLM-5.3	95.4%	59.2%	estimated ± 2.4 pp, medium confidence
GLM-5.3-Flash	92.0%	52.7%	estimated ± 2.4 pp, low confidence

Fitted on

Model	Vals SWE-bench	FrontierCode 1.1 Extended
Claude Opus 5	97.0%	63.6%
GPT-5.6 Luna	93.0%	55.1%
GPT-5.6 Sol	96.2%	60.6%
GPT-5.6 Terra	95.4%	55.8%
Grok 4.6	95.6%	61.3%