
So you want to use OpenRouter? — Mo Moustafa
Might seem simple on the face of it, but unfortunately it’s pain all the way down.
OpenRouter shared a ranking of large language models (LLMs) based on the percentage of JSON violations observed over the past week for structured output requests. **Qwen, Mistral, and GPT-4o-mini demonstrated the lowest rates of JSON violations, performing best in this regard.** In contrast, **DeepSeek v3 and Sonnet 4 showed poor performance, each exhibiting a JSON violation rate exceeding 20%.** This ranking highlights significant reliability differences in structured output between leading LLMs.