OpenHarmony Bench v1.0 (OpenHarmony Bench)
We show this table for reference; we do not rank on it.
An app-level coding benchmark that asks DevEco Code configurations to implement observable behavior in buildable OpenHarmony ArkTS applications.
Benchmark score on OpenHarmony Bench — October 17, 2026
We compile the OpenHarmony Bench rows from benchmark-owner or independent runs. Qwen3.8 Max leads the table at 60.8%, followed by GLM-5.3 (60.8%) and DeepSeek V4.1 Flash 1017 (60.3%). We do not use these results to rank models overall.
Qwen3.8 Max
Alibaba
GLM-5.3
Z.AI
DeepSeek V4.1 Flash 1017
DeepSeek
14 modelsCodingCurrentDisplay onlyUpdated October 17, 2026
Benchmark score table (14 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Qwen3.8 MaxAlibaba | 60.8% | Not reported | Open |
| 2 | GLM-5.3Z.AI | 60.8% | Not reported | Open |
| 3 | DeepSeek V4.1 Flash 1017DeepSeek | 60.3% | Not reported | Open |
| 4 | DeepSeek V4 Pro 0813DeepSeek | 59.0% | Not reported | Open |
| 5 | GLM-5.2Z.AI | 58.4% | Not reported | Open |
| 6 | Kimi K3Moonshot AI | 57.3% | Not reported | Pending |
| 7 | GLM-5.3-FlashZ.AI | 57.3% | Not reported | Open |
| 8 | Qwen3.8 Max PreviewAlibaba | 56.0% | Not reported | Closed |
| 9 | DeepSeek V4 Flash 0731DeepSeek | 53.8% | Not reported | Open |
| 10 | Qwen3.7 MaxAlibaba | 53.4% | Not reported | Closed |
| 11 | GLM-5.1Z.AI | 52.3% | Not reported | Open |
| 12 | Kimi K2.7 CodeMoonshot AI | 52.1% | Not reported | Open |
| 13 | DeepSeek V4 ProDeepSeek | 52.1% | Not reported | Open |
| 14 | MiniMax M3MiniMax | 48.4% | Not reported | Open |
Among the reported OpenHarmony Bench rows, Qwen3.8 Max is first at 60.8%. The third row is 0.5 points behind. The broader top-10 range is 7.4 points, so many of the published results sit in a relatively narrow band.
14 models have been evaluated on OpenHarmony Bench. The benchmark falls in the Coding category. OpenHarmony Bench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About OpenHarmony Bench
Year
2026
Tasks
153 app-development and bug-fix tasks
Format
Task completion through DevEco Code
Difficulty
End-to-end OpenHarmony application development
The v1.0 snapshot contains 153 top-level tasks and 242 executable feature checks across 42 repositories. Each leaderboard row averages three full-suite DevEco Code runs. We store Task Completion as display-only model-plus-agent evidence; build rates, duration, token use, and run ranges remain in the source snapshot.
Freshness and provenance
Version
OpenHarmony Bench 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does OpenHarmony Bench measure?
An app-level coding benchmark that asks DevEco Code configurations to implement observable behavior in buildable OpenHarmony ArkTS applications.
Which model scores highest on OpenHarmony Bench?
Qwen3.8 Max by Alibaba currently leads with a score of 60.8% on OpenHarmony Bench.
How many models are evaluated on OpenHarmony Bench?
14 AI models have published results on OpenHarmony Bench in the BenchLM catalog.
Compare top models on OpenHarmony Bench
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 5,600+ readers.
One email each week. Unsubscribe anytime.