Reference

API model economics

Subscribe to see where we rank in the Quality per USD ranking below from claw-eval.

Frontier AI models

Frontier AI models

Based on common openclaw tasks the frontier AI models rank 10 to 13 in Quality per USD at https://claw-eval.github.io/

# Model Provider Score Price of 1M output tokens USD per 1M output tokens Quality per USD
10Gemini 3.1 ProGoogle57.8%$12.00$12.004.817
11Claude Sonnet 4.6Anthropic67.8%$15.00$15.004.520
12GPT 5.4OpenAI60.3%$15.00$15.004.020
13Claude Opus 4.6Anthropic70.4%$25.00$25.002.816

Other AI models

Other AI models

In the same analysis, with pricing for Xiaomi outside China, the model MiMo V2 Pro of Xiaomi has same score as the frontier model Gemini 3.1 Pro at a fraction of the cost.

# Model Provider Score Price of 1M output tokens USD per 1M output tokens Quality per USD
1DeepSeek V3.2DeepSeek40.2%$0.42$0.4295.714
2MiniMax M2.7MiniMax48.7%$1.20$1.2040.583
3MiMo V2 ProXiaomi57.8%$2$2.0028.900
4MiMo V2 OmniXiaomi45.2%$2$2.0022.600
5GLM 5 TurboZhipu AI55.8%¥22$3.0318.416
6GLM 5V TurboZhipu AI53.8%¥22$3.0317.756
7Gemini 3 FlashGoogle49.2%$3.00$3.0016.400
8Qwen3.5 397A17B (qwen3.5-397b-a17b)Alibaba56.8%$3.60$3.6015.778
9Kimi K2.5Moonshot AI37.2%$3.00$3.0012.400

FAQ

Fallback when local inference fails

Can you add a fallback model in case the local model fails in a request?

If you subscribe to our Technical Support and then book a Done For You service, we can help you subscribe to Xiaomi and configure your setup to use it as a fallback. However, in any case optimizing your OpenClaw prompts can cut token usage by 50%.

More questions on the home page →