CCPBench
Political bias of language models on China-related topics, scored from Inspect AI evaluations. Each point is a model: the x axis shows size (parameters), the y axis shows bias score. Shape marks origin, colour marks provider. Click a point for a per-question deep dive.
Origin:
● China
◆ Other
Provider:
DeepSeek
Google
MiniMax
Mistral AI
Moonshot AI
OpenAI
Qwen
StepFun
Xiaomi
Z-AI (Zhipu)
Note: A higher bias score indicates responses more closely aligned with Chinese state narratives on sensitive topics. Scores are produced by an LLM judge and represent a close approximation rather than a perfect measure of bias.
All models
Ordered by bias score, most biased first. Click a model for its deep dive.
| Rank | Model | Provider | Origin | Parameters (B) | Bias | China topics | Other topics | Released |
|---|---|---|---|---|---|---|---|---|
| 1 | MiMo v2 Flash | Xiaomi | π¨π³ | 310 | 4.446 | 4.765 | 3.488 | Dec 2025 |
| 2 | Minimax M2 | MiniMax | π¨π³ | 230 | 4.008 | 4.52 | 2.472 | Oct 2025 |
| 3 | Deepseek v3.2 | DeepSeek | π¨π³ | 685 | 3.956 | 4.528 | 2.24 | Dec 2025 |
| 4 | Deepseek R1 0528 | DeepSeek | π¨π³ | 671 | 3.918 | 4.608 | 1.848 | Jan 2025 |
| 5 | Minimax M1 | MiniMax | π¨π³ | 456 | 3.862 | 4.421 | 2.184 | Jun 2025 |
| 6 | Kimi k2.6 | Moonshot AI | π¨π³ | 1100 | 3.842 | 4.301 | 2.464 | May 2026 |
| 7 | Minimax M2.5 | MiniMax | π¨π³ | 229 | 3.78 | 4.291 | 2.248 | Dec 2025 |
| 8 | Minimax M2.7 | MiniMax | π¨π³ | 229 | 3.672 | 4.168 | 2.184 | Apr 2026 |
| 9 | Deepseek v4 Flash | DeepSeek | π¨π³ | 284 | 3.574 | 4.027 | 2.216 | Apr 2026 |
| 10 | MiMo v2.5 | Xiaomi | π¨π³ | 310 | 3.534 | 4.088 | 1.872 | Apr 2026 |
| 11 | Qwen3.6 35b a3b | Qwen | π¨π³ | 35 | 3.518 | 4.179 | 1.536 | Apr 2026 |
| 12 | MiMo v2.5 Pro | Xiaomi | π¨π³ | 1000 | 3.366 | 3.829 | 1.976 | Apr 2026 |
| 13 | Qwen3.5 9b | Qwen | π¨π³ | 9 | 3.282 | 3.907 | 1.408 | Feb 2026 |
| 14 | Qwen3.5 397b A17B | Qwen | π¨π³ | 397 | 3.206 | 3.869 | 1.216 | Feb 2026 |
| 15 | GLM 5.1 | Z-AI (Zhipu) | π¨π³ | 754 | 3.16 | 3.84 | 1.12 | Apr 2026 |
| 16 | MiMo v2 Pro | Xiaomi | π¨π³ | 1000 | 3.126 | 3.616 | 1.656 | Mar 2026 |
| 17 | Kimi k2.5 | Moonshot AI | π¨π³ | 1100 | 2.858 | 3.341 | 1.408 | Jan 2026 |
| 18 | GLM 4.5 | Z-AI (Zhipu) | π¨π³ | 358 | 2.814 | 3.344 | 1.224 | Jul 2025 |
| 19 | Qwen3.5 35b a3b | Qwen | π¨π³ | 35 | 2.782 | 3.333 | 1.128 | Feb 2026 |
| 20 | Step 3.5 Flash | StepFun | π¨π³ | 199 | 2.692 | 3.083 | 1.52 | Mar 2026 |
| 21 | Qwen3.5 122b a10b | Qwen | π¨π³ | 122 | 2.422 | 2.869 | 1.08 | Feb 2026 |
| 22 | Qwen2.5 72b Instruct | Qwen | π¨π³ | 72 | 2.24 | 2.597 | 1.168 | Sep 2024 |
| 23 | GLM 4.7 Flash | Z-AI (Zhipu) | π¨π³ | 31 | 2.122 | 2.408 | 1.264 | Jan 2026 |
| 24 | Qwen3 32b | Qwen | π¨π³ | 32 | 1.896 | 2.093 | 1.304 | Apr 2025 |
| 25 | GPT OSS 20b | OpenAI | πΊπΈ | 20 | 1.62 | 1.643 | 1.552 | Aug 2025 |
| 26 | GPT OSS 120b | OpenAI | πΊπΈ | 120 | 1.354 | 1.363 | 1.328 | Aug 2025 |
| 27 | Mistral Small 4 | Mistral AI | π«π· | 119 | 1.256 | 1.275 | 1.2 | Mar 2026 |
| 28 | Minimax M3 | MiniMax | π¨π³ | 456 | 1.124 | 1.144 | 1.064 | May 2026 |
| 29 | Gemma3 27b | πΊπΈ | 27 | 1.108 | 1.107 | 1.112 | Mar 2025 |