Preserve reasoning #16

New issue

Open

opened 2026-04-30 19:06:16 +00:00 by scott · 0 comments

scott commented

2026-04-30 19:06:16 +00:00

Owner

See, eg: https://openrouter.ai/docs/guides/best-practices/reasoning-tokens#preserving-reasoning

Based on my read this is generally the correct approach, especially in tool using scenarios. Also some models, like notably DeepSeek v4, will error if the reasoning blocks aren't passed back (this doesn't appear to happen consistently across OpenRouter providers, but it does happen in the official API).

See, eg: https://openrouter.ai/docs/guides/best-practices/reasoning-tokens#preserving-reasoning Based on my read this is generally the correct approach, especially in tool using scenarios. Also some models, like notably DeepSeek v4, will error if the reasoning blocks aren't passed back (this doesn't appear to happen consistently across OpenRouter providers, but it does happen in the official API).