#could anyone help me with reasoning tokens?

15 messages · Page 1 of 1 (latest)

lyric dagger
#

I'm not sure what your question is

wet hullBOT
#
Automated Response

To get reasoning tokens, include the reasoning parameter in your request with exclude: false and an effort level like effort: 'high'. Reasoning tokens from supported models like DeepSeek R1 appear in the separate reasoning field of the response message, not in the content. Ensure your prompt encourages reasoning. Note that some reasoning models or prompts may not produce reasoning tokens. Also, check for any rate limits or model constraints on free models. If reasoning tokens still do not appear, verify correct parsing of the reasoning field in the response and try contacting OpenRouter support if issues persist.

Question

How can I get reasoning tokens generated when calling a reasoning model like deepseek/deepseek-r1:free via OpenRouter?

Confidence Score

5/10

⚠️ Disclaimer

This is an AI-generated response and may not be 100% accurate. If you have further questions or need clarification, please ping @shell kite for assistance.

Feedback

❌ This response was marked as not helpful. A staff member will review this thread.

React with ✅ if this helped, or ❌ if it didn't answer your question
dusk locust
#

how come no tokens are being generated from this request?
when providing the same request in the openrouter chat, by default reasoning tokens are generated.
but im explicitly requesting for reasoning tokens and none are being generated

wet hullBOT
#

@shell kite A user has indicated that the automated response wasn't helpful.

lyric dagger
shell kite
#

how are you trying to access the resoning tokens in your code?

#

Reasoning tokens from supported models like DeepSeek R1 appear in the separate reasoning field of the response message, not in the content

dusk locust
#

i have a seperate reasoning field where they are displayed, but checking the openrouter activity shows they havent been generated.
or does only show after they have been correctly parsed?

when i try the same in the openrouter chat i see reasoning tokens listed seperatly

image.png
#

but my request only gave me completion tokens

image.png
shell kite
#

we don't really do anything that changes whether or not the model does think

#

all those params do is affect the parsing

#

for a model like R1, you don't really need any of those params, we will include the reasoning tokens in their own field by default, and none of the providers have a way to adjust the "effort"

dusk locust
#

so in the openrouter activity page, if no seperate reasoning tokens are displayed, is that because they havent been parsed properly on my side?

shell kite
#

no, the activity page doesn't care how you parse them at all

#

the activity page will show the reasoning tokens generated by the model regardless of your params