#websockets

Why it matters: faster models stop feeling fast if orchestration overhead eats the gain. OpenAI says WebSocket mode made agent workflows up to 40% faster end to end, while lifting effective inference speed from about 65 to nearly 1,000 tokens per second.

The bottleneck moved from GPUs to the API layer, and OpenAI changed the transport to keep up. By adding WebSocket mode and connection-scoped caching to the Responses API, the company says agentic workflows improved by up to 40% end-to-end and GPT-5.3-Codex-Spark reached 1,000 tokens per second with bursts up to 4,000.