Skip to content

PostTrain Bench

A software-engineering benchmark for post-training infrastructure and implementation tasks, evaluated through the official Harbor implementation.

3 models

Higher is better.

  1. 1GLM-5.3Z.AI39.8
  2. 2Kimi K3Moonshot AI36.6
  3. 3Hy4 previewTencent35.6
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Harbor agent evaluation
Tasks
Post-training software-engineering tasks
Authors
Moonshot AI (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.