Frontiercode

Grades scope creep, large diffs and unhelpful tests, not just passing tests. Even the best model passes 58%, so merge-ready output is still unreliable.

• articleBenchmarking Gpt 6 Astra
• articleArtificial Analysis Capability Indices V1 1
• articleAa 3d
• articleClaude Haiku 5 5

• postFrontierCode 1.1 leaderboard grades coding models on merge-readinessArtificialAnlys
• postFrontierCode 1.1 benchmarks mergeable PRs on Artificial AnalysisCognition
• postDevin Adds Claude Opus 5.5, Tops FrontierCode BenchmarkCognition

Checking sign-in…

Loading comments…