OpenHarmony Bench: leaderboard
Coding agents complete OpenHarmony software-development tasks under the paper’s evaluation harness. The original paper reports eight models over 153 tasks, with separate new-feature, specification-driven and bug-fix populations. Scores are mean task-completion percentages; interval endpoints are not treated as additional scores. This board reports OpenHarmony Bench using Mean Task Completion (%).
Metric: Mean Task Completion (%). Source: arxiv.org. Saturation forecast: Around July 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.2 | 58.39 | #80 |
| 2 | Kimi K3 | 57.3 | #35 |
| 3 | Qwen 3.7 Max | 53.38 | #85 |
| 4 | GLM-5.1 | 52.29 | #109 |
| 5 | DeepSeek V4 Pro | 52.07 | #112 |
| 6 | Kimi K2.7 Code | 52.07 | #95 |
| 7 | MiniMax-M3 | 48.36 | #151 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openharmony-bench · How It Works · Data refreshed daily, snapshot 2026-11-05.