benchgap
Coding

OpenHarmony Bench v1.2 leaderboard

As of 2026-11-21, the highest measured score on OpenHarmony Bench v1.2 is 71.3% by DeepSeek V4.1 Flash 1017. 23 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1DeepSeek V4.1 Flash 101771.3%measured
2Qwen3.8 Max67.8%measured
3Claude Fable 5.163.8%estimated ± 11.0 pp, low confidence
4Claude Fable 5.263.8%estimated ± 11.0 pp, low confidence
5Claude Opus 5.563.8%estimated ± 11.0 pp, low confidence
6Claude Sonnet 5.563.8%estimated ± 11.0 pp, low confidence
7Gemini 4 Argon63.8%estimated ± 11.0 pp, low confidence
8Gemini 4 Carbon63.8%estimated ± 11.0 pp, low confidence
9GPT-5.6 Sol63.8%estimated ± 11.0 pp, low confidence
10GPT-6.1 Sol63.8%estimated ± 11.0 pp, low confidence
11GPT-6.2 Sol63.8%estimated ± 11.0 pp, low confidence
12GPT-6 Astra63.8%estimated ± 11.0 pp, low confidence
13Muse Wonder Full63.8%estimated ± 11.0 pp, low confidence
14Gemini 4 Hydrogen63.8%estimated ± 11.0 pp, low confidence
15Muse Spark 1.363.8%estimated ± 11.0 pp, low confidence
16Muse Wonder63.8%estimated ± 11.0 pp, low confidence
17Grok 4.763.8%estimated ± 11.0 pp, low confidence
18Claude Opus 563.8%estimated ± 11.0 pp, low confidence
19Grok 4.663.8%estimated ± 11.0 pp, low confidence
20MiMo-V2.6-Pro63.8%estimated ± 11.0 pp, low confidence
21Claude Haiku 5.563.8%estimated ± 11.0 pp, low confidence
22Kimi K363.7%estimated ± 11.0 pp, low confidence
23GLM-5.360.7%measured
24GPT-6 Luna55.5%estimated ± 11.0 pp, low confidence
25Gemini 3.8 Flash52.8%estimated ± 11.0 pp, low confidence
26Kimi K3.1 Pro52.4%measured
27GLM-5.3-Flash50.4%measured
28Qwen3.8-27B45.7%estimated ± 11.0 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General