LiveBench Coding 2.0

Fix real bugs from merged pull requests in a large frozen codebase. Hidden tests grade the fix; a regression scores 0.

ModelsModels with a result.52
Top resultBest result on this benchmark.98.3%Claude Fable 5
Top-3 spreadPoints from first to third.2.0 pts
UpdatedDate the source changed the results.14 Oct 2026

Result and cost

52 of 52 results

Results

52 results
#ModelResultThe score from the source.CostUS dollars to run one task.
09GPT-6 Solmax effort93.1%$0.977
28GLM-5.3max effort82.8%$2.14
44Inkling64.0%$1.88