[GH-ISSUE #8696] DeepSeek-R1 support #15220

New Issue

GiteaMirror · 2026-04-19T21:29:07-05:00

GiteaMirror commented

2026-04-19 21:29:07 -05:00

Originally created by @matt23654 on GitHub (Jan 20, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/8696

Is your feature request related to a problem? Please describe.
DeepSeek-R1 API returns its thoughts inside the reasoning_content parameter. Currently this is ignored by OpenWebUI.

Describe the solution you'd like
Display the reasoning/thoughts for reasoning models like DeepSeek-R1.

Describe alternatives you've considered
Their API approach, of return "reasoning_content" for the long-cot but then to only pass "content" (the usual short reply) for multi-turn seems like a very good token conserving way to provide APIs for reasoning models. I would expect multiple providers to adopt this approach in the future.

Additional context
API docs are here: https://api-docs.deepseek.com/guides/reasoning_model

Originally created by @matt23654 on GitHub (Jan 20, 2025). Original GitHub issue: https://github.com/open-webui/open-webui/issues/8696 **Is your feature request related to a problem? Please describe.** DeepSeek-R1 API returns its thoughts inside the ``reasoning_content`` parameter. Currently this is ignored by OpenWebUI. **Describe the solution you'd like** Display the reasoning/thoughts for reasoning models like DeepSeek-R1. **Describe alternatives you've considered** Their API approach, of return "reasoning_content" for the long-cot but then to only pass "content" (the usual short reply) for multi-turn seems like a very good token conserving way to provide APIs for reasoning models. I would expect multiple providers to adopt this approach in the future. **Additional context** API docs are here: https://api-docs.deepseek.com/guides/reasoning_model

GiteaMirror closed this issue

2026-04-19 21:29:08 -05:00

GiteaMirror commented

2026-04-19 21:29:09 -05:00

@gkopiev commented on GitHub (Jan 20, 2025):

Also there is no option in Open WebUI (or maybe I haven't found it) to disable sending certain parameters for deepseek-reasoner (DeepSeek R1) model. It leads to errors like that:

400: deepseek-reasoner does not support the parameter temperature

Documentation mentions that top_p parameter is also not supported, but I don't receive message about that when trying to use deepseek-reasoner.

@gkopiev commented on GitHub (Jan 20, 2025): Also there is no option in Open WebUI (or maybe I haven't found it) to disable sending certain parameters for deepseek-reasoner (DeepSeek R1) model. It leads to errors like that: > 400: deepseek-reasoner does not support the parameter `temperature` [Documentation](https://api-docs.deepseek.com/guides/reasoning_model) mentions that top_p parameter is also not supported, but I don't receive message about that when trying to use deepseek-reasoner.