OpenAI has introduced a WebSocket-based execution mode for its responses API to improve the performance of agentic workflows used in coding agents and real-time AI systems. The change replaces the traditional HTTP request-response pattern with a persistent, bidirectional connection between client and server, targeting latency and coordination overhead in multi-step reasoning workflows.

What Happened

In April 2026, OpenAI announced that it had integrated WebSockets into its Responses API, reducing latency by 40% and enabling faster agentic workflows. The company reported that early production use showed up to 40% latency reduction and improved throughput in high-concurrency scenarios.

The update addresses a growing bottleneck in agentic systems where each step in a workflow, such as tool calls, intermediate reasoning, and follow-up queries, previously required separate HTTP requests. As inference speeds improved, these repeated network round-trip times became a dominant source of latency and operational complexity.

Background and Context

The introduction of WebSockets into the Responses API is a significant development in the field of agentic AI. Agentic workflows involve multiple steps, including tool calls, intermediate reasoning, and follow-up queries, which can lead to repeated network round-trip times and latency.

OpenAI's GPT-5.3-Codex-Spark model achieved speeds of over 1,000 tokens per second (TPS) on specialized Cerebras hardware, making the inefficiencies in the API more glaring. Each agentic workflow involved multiple back-and-forth API calls, processing redundant context and conversation history, which compounded latency as tasks grew more complex.

OpenAI began by optimizing critical-path latency for single requests, including caching rendered tokens and model configurations in memory, reducing network hops by bypassing intermediate services, and streamlining safety classifiers to flag issues faster. However, these optimizations were insufficient to fully leverage the speed of GPT-5.3-Codex-Spark.

Why It Matters

The introduction of WebSockets into the Responses API has significant implications for the industry. By reducing latency and improving throughput in high-concurrency scenarios, OpenAI's update can help improve the performance of agentic workflows used in coding agents and real-time AI systems.

The change reflects a broader focus on the transport layer in agentic systems, where communication patterns and connection management influence overall performance. As Ofek Shaked, a Vibe coder, noted, "WebSockets for agent state is such an obvious but huge win. No more cold starts killing your multi-tool chains."

What Comes Next

The introduction of WebSockets into the Responses API is a significant development in the field of agentic AI. As the industry continues to evolve, it will be interesting to see how other companies respond to this update and whether they adopt similar approaches to improving performance.

Key Facts

  • OpenAI has introduced a WebSocket-based execution mode for its responses API to improve the performance of agentic workflows.
  • The change replaces the traditional HTTP request-response pattern with a persistent, bidirectional connection between client and server.
  • Early production use showed up to 40% latency reduction and improved throughput in high-concurrency scenarios.
  • The update addresses a growing bottleneck in agentic systems where each step in a workflow requires separate HTTP requests.
  • OpenAI's GPT-5.3-Codex-Spark model achieved speeds of over 1,000 tokens per second (TPS) on specialized Cerebras hardware.
  • The introduction of WebSockets into the Responses API reflects a broader focus on the transport layer in agentic systems.