← ChatGPT & Codex changelog

OpenAI API — September, 2026

This release features the launch of GPT-6 Astra, our most capable model for complex end-to-end work, complete with async tool calling and mid-turn steering. New capabilities include the Agents API public beta for building managed agents, GPT-Live 1 for full-duplex voice conversations, and GPT Image 2.5 Sunburst and Flare for advanced image generation. Developers also benefit from project API key expiration dates, general availability of Prompt Cache Diagnostics, and improved API error handling, enhancing platform control and performance.

Added

  • Released the Agents API in public beta. Build agents with a managed Codex harness while OpenAI handles session orchestration, context compaction, and recovery.
  • Released GPT Image 2.5 Sunburst and GPT Image 2.5 Flare for image generation and editing through the Image API and the Responses API image generation tool.
  • Use Sunburst for workflows where editing precision matters most, or Flare for fast, high-quality everyday image generation. Both models support the new xhigh and max quality settings and use GPT Image 2 token rates. See the image generation guide and pricing.
  • Released GPT-6 Astra, our most capable model, built for the hardest end-to-end work.
  • Added new controls for long-running work with GPT-6 Astra in the Responses API:

Changed

  • You can now set expiration dates when creating project API keys. Administrators can also enforce a maximum key lifetime at the organization or project level in Platform settings, requiring newly created keys to expire within the configured limit. See production best practices for guidance on key expiration and rotation.
  • Use durable sessions to continue work across turns, stream progress, and connect your own tools and MCP servers. Run agents in OpenAI-hosted sandboxes or connect a sandbox from your own infrastructure or a supported provider.
  • Start with the Agents API quickstart.
  • GPT-Live 1 is now generally available in the API. Build full-duplex voice conversations that can continue while a backend model or agent handles reasoning and tools.
  • Use Responses delegation with an OpenAI model, or client delegation to connect your own backend. Voice sessions cost $0.05 per minute, billed per second; backend model and tool usage is charged separately.
  • Start with GPT-Live, prompting, and migration guidance. See pricing for details.
  • Prompt Cache Diagnostics is now generally available in the Responses API for GPT-5.6 and later supported models.
  • Compare cache reuse against a previous response, identify reasons for cache misses, and follow troubleshooting guidance to improve cache reuse.
  • GPT-Rosalind (gpt-rosalind-research) is now generally available through the trusted-access program for approved internal life sciences research.
  • Standard pricing is $5 per 1M input tokens, $0.50 per 1M cached input tokens, and $25 per 1M output tokens. Billing begins on October 5, 2026. See pricing for details.
  • Use GPT-6 Astra for reasoning, coding, computer use, research, and document creation. It combines these capabilities to carry complex tasks from an initial request to a finished result, using the context and tools you provide.
  • Key changes to consider when migrating:
  • GPT-6 Astra does not support the none reasoning effort level.
  • GPT-6 Astra does not support custom temperature or top_p values or log probabilities (logprobs).
  • Tool calling requires the Responses API. If you use tools with Chat Completions, follow the Responses migration guide.
  • Misalignment monitoring asynchronously checks for potential issues during agent work in supported Responses API requests. Checks can trigger safety alerts or stop a conversation for review.
  • Start with Using GPT-6 Astra for capabilities, prompting, and migration guidance. Explore computer use for browser and desktop workflows, and see pricing for available inference tiers.
  • Async tool calling: Let the model continue working while your application runs function or custom tools, then return results as they become available.
  • Mid-turn steering: Send additional instructions while a response is in progress over WebSockets, so the model can incorporate corrections or changing requirements.
  • Change reasoning effort mid-conversation: Increase effort for difficult work or reduce it for routine follow-ups while preserving the cached prompt prefix.
  • Updated API errors so applications can distinguish traffic that increases too quickly from temporary model overload.
  • Traffic that increases too quickly can return a 429 error with the slow_down code. Temporary model overload returns a 503 error with the server_is_overloaded code. Both responses may include Retry-After. When the header is present, wait at least as long as it specifies before retrying. If it's missing, use exponential backoff. See the error codes guide and rate limits guide.
  • Connections to api.openai.com can now use IPv6.