Help understanding the differences between R1 and R1 1776

[warped coyote]

In requests to R1 in normal chat prompts, it seems that I don't usually get thinking responses. WIth R1 1776 I get thinking in the responses, but they are not enclosed in any think tags. Is the R1 1776 model actually behaving differently, or is this a configuration difference on your end?

[charred pelican]

The R1 1776 model has a different chat template, meaning they don't return <think> tags, and therefore we can't parse the reasoning out

At least, that's how it's configured from Perplexity, a different provider could use a different chat template in the future

I tried to add this as a note on our model page here, lmk if this makes sense https://openrouter.ai/perplexity/r1-1776

Note: As this model does not return <think> tags, thoughts will be streamed by default directly to the content field.

[warped coyote]

thanks yeah, i tested with their api, it's not reliably returning <think> opening tags but sometimes I get </think> closing tags .. i think they just messed up a little

thanks for the response!

[charred pelican]

of course, any time

[warped coyote]

oh btw is there a way to keep the think tags intact with other providers if i want theM?

[charred pelican]

we parse those out and put them all in the reasoning field of the completion object

if you set include_reasoning to true

[warped coyote]

thanks!

[charred pelican]

Yep np, docs here https://openrouter.ai/docs/api-reference/parameters#include-reasoning

[warped coyote]

seems they fixed their api btw

[brisk crest]

Hmm, maybe it's possible for OR to parse the <think> tags now and put that stuff in .reasoning?

Ah it looks like it already does. Nice!