I keep trying to use the Sonar Reasoning model but sometimes end up with no usable result because it cuts for length at 32k output tokens while it’s still reasoning. The documentation makes me think it should be around 128k given it advertises a 128K context length
I tried setting “max_tokens”:100000 but it’s still cutting off at 32k. Are there default and max values for each model type posted somewhere? I’m able to get outputs consistently when using the web UI with the same prompt. Do the API end points have lower cut offs than the web UI? Is there another way I should be trying to configure the max_token value?