Scira AISScira AI
Jul 30

Follow-up: DeepSeek v4 Pro and GLM 5.2 still capping at ~4,096 output tokens

Thanks for moving this to Completed. I wanted to check in on it — I haven't been able to confirm any change in behavior yet, and I'm still seeing outputs cut off at around 4,096 tokens, including mid-inference during tool step execution on DeepSeek v4 Pro and GLM 5.2.Could you let me know whether this is still on the team's radar, and if so, whether there's an expected timeline or any settings I can adjust on my side in the meantime? Happy to share additional reproduction details or logs if that would help.
CompletedCompleted
changed status toCompleted·last week
changed status toReviewing·3 weeks ago