gproxy v2.9.4
v2.9.4
Restores incremental OpenAI Chat streaming when a Claude upstream reports
input-token usage in its opening event.修复 Claude 上游在起始事件中报告输入 token 用量时,OpenAI Chat 下游失去
增量流式输出的问题。
English
Fixed
- Claude-to-OpenAI Chat responses stream incrementally again. The downstream cost decorator no longer mistakes an opening Chat chunk containing converted Claude input usage for a terminal usage frame and buffer the remaining response. It now holds only usage-bearing chunks that also carry a real completion signal.
简体中文
修复
- Claude 转 OpenAI Chat 恢复增量流式输出。 下游 cost 注入器不再把含有 Claude 起始输入用量的首个 Chat chunk 误判为终止 usage 帧并缓存后续全部响应;现在只有同时携带真实结束信号的 usage chunk 才会被保留等待结算。