Jul 7
Request to Review Maximum Output Context Length for DeepSeek v4 Pro and GLM 5.2
Based on my observations, DeepSeek v4 Pro and GLM 5.2 may be limited to approximately 4,096 output tokens. Additionally, the output may sometimes stop during the inference phase of a tool's step execution.
Since I primarily use these models in Japanese, token consumption may be higher than in English, so my observations could be influenced by that.
Could you please review this behavior and adjust the settings if appropriate?
In Progress