Skip to content
integration bench from HeyMilo
  1. Home
  2. Releases

15+ models evaluated

15+ models are run on the public 50, including Claude Fable 5 and GPT-5.6 Sol. One official attempt per model per task, provider harness, 60-minute wall clock, one scorer version. Claude models in Claude Code; GPT-5.6 Sol, Terra, and Luna in Codex; the rest in OpenCode.

22 August 2026

What shipped

  • Models. 15+ on the public 50, including Claude Fable 5 and GPT-5.6 Sol. Reasoning effort is the highest level the provider exposes, recorded per attempt.
  • Harness. Claude models in Claude Code. GPT-5.6 Sol, Terra, and Luna in Codex. The rest in OpenCode.
  • Runs. One official attempt per model per task. 60-minute wall clock. Same task tree, same vendor images, same scorer.
  • Metric. Mean task score over all 50 public tasks. Scorer: task-score-v4-mandatory-gated. Required checks gate the task; remaining checks award credit after that.

The write-up is How we built Integration Bench.