Model list
Adopted from OpenResponses on 2026-10-01, at Gunnar’s request, so that both apps offer the same models with the
same settings. OpenCone carries a copy of OpenResponses’ built-in list and the code that reads it (ported files name
their source at the top). Keep the two copies in step: when OpenResponses changes its list, copy
OpenResponses/Resources/ModelCatalog/ModelCatalog.json over OpenCone/Resources/ModelCatalog/ModelCatalog.json,
keeping OpenCone’s notes line.
| Source | What it gives | When |
|---|---|---|
OpenCone/Resources/ModelCatalog/ModelCatalog.json, built in | The models the app knows, in menu order, with their reasoning efforts and summaries; the default model; retired models and where a saved one moves | Always |
| GET /models with the person’s OpenAI key | Which models the account can use, and each one’s shutdown_date | Once per launch (SettingsViewModel.refreshAccountModels) |
The model’s page on OpenAI’s docs site, https://developers.openai.com/api/docs/models/<id>.md | Reasoning efforts, a summary, and whether the Responses API supports it, for a model the built-in list does not name | The first time the account lists the model, again after 7 days, at most 3 pages per launch |
A new general-purpose model (for example gpt-6.2-sol) appears at the top of the model menu at the next launch after
the account lists it. Until its page has been read, its reasoning options start at low (GPT-6 Astra and GPT-6.1 Sol
reject none with HTTP 400). A page that says the Responses API does not support the model takes it out of the menu.
A model whose GET /models entry has a shutdown_date within 30 days, or past, counts as retired: it leaves the model
menu, and a saved choice of it moves to its replacement when Settings loads or the account list is refreshed. A
retired model with no replacement of its own follows the longest retired entry that has one (o1-mini follows o1
to gpt-5.6-sol), or moves to defaultModel when none matches.
What OpenCone uses from it
| Piece | Where |
|---|---|
| Default model | Configuration.completionModel, which is CurrentModelCatalog.defaultModel |
| Model menu | SettingsViewModel.availableCompletionModels, CurrentModelCatalog.selectionModels |
| Reasoning efforts offered, and the effort sent | CurrentModelCatalog.reasoningEfforts(for:) and normalizedEffort(_:model:), applied in Settings and to routing calls |
| Retired models | CurrentModelCatalog.isRetired and replacement(for:), applied when Settings loads |
| Background model | CurrentModelCatalog.utilityModel (gpt-6-luna) drafts each index’s one-line summary |
OpenResponses also reads pro (reasoning mode) and asyncTools from the list; OpenCone keeps those fields so the
file stays identical, but does not send either.
The file’s format, its validation (ModelCatalog.problem()) and its upkeep are described in OpenResponses’
docs/model-catalog.md.