MobileKernelBench: leaderboard
Metric: Functional correctness rate (%): share of MobileKernelBench's 190 operator tasks (95 ONNX primitive operators in 12 categories), the model writes an MNN CPU kernel from one standardized prompt, compiled and run on a Xiaomi 13 phone (Snapdragon 8 Gen 2), whose output matches the ONNX reference; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 34.2 | #138 |
| 2 | GPT-5 | 33.2 | #91 |
| 3 | Gemini 2.5 Flash | 15.8 | #237 |
| 4 | Qwen 3 235B A22B 2507 (Thinking) | 13.7 | #253 (Qwen 3 235B A22B 2507) |
| 5 | DeepSeek R1 0528 | 10 | #217 |
| 6 | Llama 3.1 405B Instruct | 6.3 | #447 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=mobilekernelbench · How It Works · Data refreshed daily, snapshot 2026-10-11.