OpenResponses 2.7 release notes
Prepared September 24, 2026. The App Store What’s New text is release_notes.txt; the full list of changes is in the changelog.
Models
| Model | Reasoning effort | Price per 1M tokens (input / output) | Role in the app |
|---|---|---|---|
gpt-6-sol | none, low, medium (default), high, xhigh, max | $2 / $10 | Default for new presets and the Workbench |
gpt-6-astra | low, medium, high, xhigh, max | $10 / $50 | Pro reasoning mode, async tool calls |
gpt-6-luna | none, low, medium (default), high, xhigh, max | $0.10 / $0.50 | MCP tool discovery |
gpt-image-2.5-flare | quality auto, low, medium, high, xhigh, max | $5 text in / $30 image out | Default image model |
gpt-image-2.5-sunburst | quality auto, low, medium, high, xhigh, max | $5 text in / $30 image out | Selectable |
Source: OpenAI model pages and changelog, fetched September 24, 2026. Prices are OpenAI’s list prices and are billed to the user’s own API account.
Retirements the app now accounts for
| Retirement | Shutdown | App behavior |
|---|---|---|
gpt-5, gpt-5-mini, gpt-5-nano, o3 snapshots | December 11, 2026 | Left out of the offline list |
gpt-realtime, gpt-realtime-mini | January 20, 2027 | Saved voice setting moves to gpt-realtime-2.1 |
whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe | February 26, 2027 | Not used; voice transcription already uses gpt-live-transcribe |
Reusable prompts API (v1/prompts) | November 30, 2026 | Removed: published-prompt fields and the prompt request object |
| Fine-tuning job creation for most organizations | January 6, 2027 | Removed: fine-tuning screen, service and dataset tools |
| Assistants API | August 26, 2026 (shut down) | Removed: service, models and the migration lab |
web_search_preview, computer_use_preview, computer-use-preview | Retired | Removed; saved configurations decode to web_search and computer |
| o1, o1-pro, o3-mini, o4-mini, gpt-4, gpt-4-turbo, gpt-4.1-nano, gpt-3.5-turbo | October 23, 2026 | Hidden; presets move to the documented replacement |
Current API coverage
| Area | Now in the app |
|---|---|
| Voice | GPT Realtime 2.1 (Realtime API) and GPT-Live 1 (Live API), chosen in Settings → Model → Voice |
| Responses parameters | moderation, external_web_access, container memory_limit, input_fidelity, output_compression, hybrid_search, service tiers through ultrafast |
| API Workbench | Responses, conversations, models, moderations, embeddings, files, vector stores, containers, batches, Realtime client secrets, voice consents |
Sources: OpenAI API reference Markdown exports (/api/reference/llms.txt and resource pages), downloaded September 24, 2026.
Forward compatibility
Later general-purpose GPT releases are recognized by version number, so a model such as gpt-6.1-sol appears in the model list with current reasoning and tool controls once the account’s model list includes it. Reasoning effort follows the GPT-6 pattern: every level for most models, and no none for Astra-class models. If a future model rejects a setting, the API returns an error message naming it, and the model ID field and API Workbench remain available.