The tool uses SSE (Server-Sent Events) technology to realize true real-time streaming, and each token is pushed to the client immediately after generation. Performance test data shows that when generating a text of 1000 tokens, the time to first byte arrival (TTFB) is only 50ms, which is 8 times faster than conventional APIs. The streaming API design consists of two layers: the base layer returns according to the OpenAI standarddelta.contentThe reinforcement layer is passed through thedelta.reasoning_contentExposing Gemini's real-time reasoning process. In a dialog bot case, this mechanism reduces the user's waiting perception time by 761 TP3T, while supporting an intermediate result intervention feature that allows the user to correct the generation direction in real time.
This answer comes from the articlegeminicli2api: Proxy tool to convert Gemini CLI to OpenAI-compatible APIsThe
