API Changelog

GPT-Live (gpt-live-1 and gpt-6-luna) now has native Decisions integration, letting apps act while the user is speaking or react to Live output. Live can invoke Decisions through native delegation, or decisions can be triggered from input/output transcript deltas.

Responses API users can set service_tier: "ultrafast" on gpt-6.1-sol to reduce time between generated output tokens. Available to all API users subject to rate limits, with global processing and US and EU data residency.

The chat-latest snapshot now points to the latest model available in ChatGPT for Plus, Pro, Business, and Enterprise users, with the underlying snapshot updated regularly. Production API usage should use the GPT-6 model family; chat-latest is for testing the latest chat improvements.

API usage tiers are simplified from five to three: Build, Launch, and Grow. Organizations automatically upgrade as total credit purchases reach tier minimums.

Released the Decisions API in beta with gpt-6-luna, turning text and images into typed answers 10x faster than the Responses API.

Admins of eligible organizations can now accept the standard Business Associate Agreement and enable HIPAA compliance support directly from Organization settings > General.

Ultrafast mode for GPT-6 Astra is available in the Responses API via service_tier: "ultrafast", reducing the time between generated output tokens. It is subject to rate limits, restricted to global processing and US data residency, and EU and other regional inference residency aren't supported.

Computer use is now available in the Agents API, letting agents complete tasks in an OpenAI-hosted browser, with website access approvals and sign-in handled by the calling application.

GPT-6.1 Sol (gpt-6.1-sol) is available for complex coding and professional work at a lower cost than GPT-6 Astra, priced at $2 input, $0.10 cached input, $2.50 cache write, and $10 output per 1M tokens for prompts up to 272K input tokens. It also supports Multi-agent in beta, letting the model delegate work to subagents in a Responses API request.

An image encoding bug that degraded image understanding in GPT-6 Sol and GPT-6 Luna is fixed, improving results on visual tasks in the API and Codex, including computer use. OpenAI recommends rerunning evaluations and retrying workflows that use image inputs.

Administrators can now restrict API key creation at the organization and project levels to service-account keys only, user-owned project keys only, or disable new key creation entirely, with organization restrictions taking precedence over project settings. Existing API keys are unaffected.

Project API keys can now be created with expiration dates, and administrators can enforce a maximum key lifetime at the organization or project level in Platform settings so newly created keys expire within the configured limit.

The Agents API is available in public beta for building agents on a managed Codex harness, with OpenAI handling session orchestration, context compaction, and recovery. It supports durable sessions, streaming progress, custom tools and MCP servers, and OpenAI-hosted or self-provided sandboxes.

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, with billing beginning October 5, 2026.

GPT Image 2.5 Sunburst and GPT Image 2.5 Flare are available for image generation and editing through the Image API and the Responses API image generation tool. Both support new xhigh and max quality settings and use GPT Image 2 token rates.

Prompt Cache Diagnostics is now generally available in the Responses API for GPT-5.6 and later supported models. It lets developers compare cache reuse against a previous response, identify reasons for cache misses, and follow troubleshooting guidance to improve cache reuse.

New Responses API controls for long-running GPT-6 Astra work let the model keep running while your app executes function or custom tools and return results as available, accept mid-turn steering over WebSockets while a response is in progress, and change reasoning effort mid-conversation while preserving the cached prompt prefix.

GPT-6 Astra is released for reasoning, coding, computer use, research, and document creation, carrying complex tasks from an initial request to a finished result. It does not support the none reasoning effort level, custom temperature or top_p values, or log probabilities, and tool calling requires the Responses API.

Rapid traffic increases now return a 429 with the slow_down code, while temporary model overload returns a 503 with server_is_overloaded, letting applications tell the two apart. Both responses may include Retry-After; wait at least as long as it specifies, or use exponential backoff when absent.

Connections to api.openai.com can now use IPv6.