Leaderboard
Best local LLM for Chinese
Measured Chinese scores of quantized local models, per GPU. Highest average so far: qwen3.8-27b-q3 on AMD Radeon RX 9070 XT.
| Model | GPU | Decode tok/s | Knowledge (translated) % | Local exams % | Math % | Translation chrF++ | Source |
|---|---|---|---|---|---|---|---|
qwen3.8-27b-q3 | AMD Radeon RX 9070 XT | 50 | 76.2 | 74.7 | 84.4 | 49.1 | maintainers |
qwen3.5-9b-q4 | AMD Radeon RX 9070 XT | 42 | 74.5 | 73.5 | 84.4 | 46.6 | maintainers |
gemma4-26b-a4b-qat | AMD Radeon RX 9070 XT | 90 | 73.5 | 66.5 | 88.8 | 47.7 | maintainers |
qwen3.8-27b-iq2 | AMD Radeon RX 9070 XT | 40 | 67.8 | 67.8 | - | - | maintainers |
qwen3.5-4b-q4 | AMD Radeon RX 9070 XT | 78 | 69.0 | 69.0 | 84.4 | 45.4 | maintainers |
gemma4-e4b-q4 | AMD Radeon RX 9070 XT | 105 | 63.5 | 53.9 | 83.6 | 45.8 | maintainers |
Zero-shot, thinking off; margins about ±4-5 points. How each
column is measured: leaderboard. Missing your GPU? Run
localllm eval --langs en,zh and share it with localllm report.