Added opt-in Windows system proxy support for authentication, including PAC, WPAD, static proxies, and bypass rules.
MCP tools now use tool search by default for improved discovery. macOS authentication can respect system proxy, PAC, and WPAD settings. Fixes include actionable Amazon Bedrock credential recovery, remote stdio MCP path support, and PowerShell AST safety checks.
Fixed an issue where multiple authorization headers were not correctly prioritized, ensuring the first header takes precedence.
Added afterCompletion hook to runTools, support for sideband call_id connections in realtime, and helpers for responses input items and stream accumulation. Also fixed structured output parsing issues including whitespace chunk breaking, incomplete output handling, and restoring output_text on streamed final response.
Configurable rollout token budgets now track usage across agent threads and abort turns when exhausted. An indexed web-search mode restricts direct page access to server-approved URLs. Fixed Linux TUI rendering after Ctrl+Z suspension and exec-server process survival through transient disconnects. Parent agents now receive terminal subagent errors instead of empty successful completions.
Added per-host personality settings with Friendly and Pragmatic options, goal editing directly in the composer, and links from forked conversations back to the original thread. Improved side chat visibility, composer autocomplete, and workspace file search.
Long pastes over 10,000 characters are now automatically converted to attachments in ChatGPT for Free and Go users, keeping the composer cleaner and preventing large pastes from consuming the context window. The threshold was also raised to 10k characters for Plus, Pro, and Business users, up from the previous 5k limit.
Enterprise and Edu workspaces can enable Slack connector actions for joining channels, creating reminders, uploading files, and updating profiles. Admins manage actions under Apps in ChatGPT; some actions require additional Slack OAuth scopes or admin approval.
ChatGPT Enterprise admins can now view credit usage trends across users, products, and models in the Global Admin Console, and set default, group-specific, or individual spend limits. Users can see their credit usage and request limit increases with context.
Updated the OpenAPI specification and generated code accordingly.
GPT-5.5 Instant now performs at a level comparable to frontier Thinking models on health evaluations, available to all free ChatGPT users. Production factuality issues flagged in health responses have fallen by 71% in the last two months, and physician-rated responses now surpass both older models and human-written responses across accuracy, communication, and completeness criteria.
Record & Replay is a new Codex app feature for macOS that lets eligible Business users demonstrate a workflow once and turn it into a reusable skill. Initial availability excludes the European Union, UK and Switzerland.
Record & Replay lets macOS users turn a demonstrated workflow into a reusable skill, initially available outside the EEA, UK, and Switzerland with Computer Use enabled. Bulk actions for automation run history, deep links for SSH connections, and improved Browser Use session persistence are also included.
Remote executors now use authenticated, end-to-end encrypted Noise relay channels, and bundled SQLite is pinned to a version containing the WAL-reset corruption fix. Plus cross-platform remote execution, per-thread executor plugin MCP servers, and auto-resolving TUI prompts.
Updated the OpenAPI specification or Stainless configuration.
Updated the OpenAPI specification or Stainless configuration.
Admin organization spend alerts are now available via the API.
Members of ChatGPT Edu workspaces without data residency configuration can now export workspace data when an admin enables the Data export setting. Exports are requested from Settings > Data controls and delivered via email as conversation JSON files.
Admin spend alerts are now available through the API.