Add the thinking version of Qwen 235B model

Open avinashkanaujiya opened this issue 4 months ago • 3 comments

It has just been released on Cerebras. Now, there are two models: thinking and non-thinking. Therefore, access to both would be really good because the thinking model would be an alternative to the 2.5 Flash model.

Aug 02 '25 06:08 avinashkanaujiya

I see that the thinking model was added and then removed. Why was this the case? @Beingpax

Aug 02 '25 16:08 avinashkanaujiya

The thinking model was too slow to respond, taking around 5 to 7 seconds. So, to prevent user confusion, I removed it.

Aug 02 '25 16:08 Beingpax

Also the output format doesn't enclose inside think tokens properly messing with our thinking token filter.

Aug 02 '25 16:08 Beingpax