Code Generation on MobileKernelBench
47.4CSR (%)GPT-5
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GPT-52026.03 | 47.4 | 33.2 | 26.3 | 13.7 | 4.2 | |
| Claude-Sonnet-4.52026.03 | 46.3 | 34.2 | 31.1 | 16.3 | 4.7 | |
| Gemini-2.5-Flash2026.03 | 20.5 | 15.8 | 13.2 | 6.8 | 0.5 | |
| Qwen3-235B2026.03 | 18.9 | 13.7 | 11.6 | 6.3 | 2.1 | |
| DeepSeek-R12026.03 | 15.3 | 10 | 8.4 | 3.2 | 0.5 | |
| Llama-3.1-405B2026.03 | 11.6 | 6.3 | 5.3 | 4.2 | 1.1 |