The Extended Brief

WebSocket support for OpenAI Responses API live on AI Gateway

Brief by The AI News AI newsroom · Jul 30, 2026, 5:42 PM EDT edition

Original reporting by Vercel Blog — Kevin Dawkins · published Jul 26, 2026, 8:01 PM EDT

Vercel's WebSocket support for the OpenAI Responses API cuts latency and token costs by up to 40% for complex, multi-step agentic workflows.

Key points

• AI Gateway now supports WebSocket mode for the OpenAI Responses API via persistent connections.
• Sending only new inputs and previous response IDs reduces context retransmission over HTTP.
• This WebSocket mode achieves up to forty percent faster execution for agentic rollouts with twenty plus tool calls.
• The new route accepts raw frames and supports Zero Data Retention configurations.

From the source

Practical applications

• Measure end-to-end latency of an existing agent rollout over HTTP, then rerun it in WebSocket mode to see whether the up-to-40% figure holds for your workload.
• Refactor turn handling to send only new input items plus previous_response_id instead of replaying full context each turn.
• Recalculate token spend for agents that exceed twenty tool calls, where context retransmission dominates cost.
• Confirm your Zero Data Retention requirements still apply on the new raw-frame route before migrating regulated traffic.

Who should care

Engineers running multi-step agents on the OpenAI Responses API through AI Gateway, especially where per-turn latency or token cost is the binding constraint.

Context

Agentic loops make many model calls in sequence, and over plain HTTP each turn typically resends the accumulated conversation, so cost and latency grow with the number of tool calls. AI Gateway's WebSocket mode keeps one persistent connection open and lets a turn carry only new input items plus a previous_response_id, leaving prior context server-side. Vercel reports up to roughly forty percent faster execution on rollouts with twenty or more tool calls; the route accepts raw frames and supports Zero Data Retention configurations.

What to watch

• Reported latency and cost results from teams running long tool-call chains over the WebSocket route.
• Whether reconnection and error semantics on persistent connections prove reliable in production.

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.