[roomote-roomote]

​Opened on behalf of @mrubens. Follow up by mentioning @roomote-roomote, in the web UI, or in Slack.

What changed

• Add GPT-6.1 Terra to the shared curated catalog for OpenAI, ChatGPT Subscription, OpenRouter/Roomote inference, Vercel AI Gateway, Azure OpenAI, Azure AI Foundry, and Requesty.
• Use Requesty's verified EU route, azure/gpt-6.1-terra@swedencentral, and Azure's gpt-6.1-terra deployment name.
• Register shared runtime model definitions for Requesty and both Azure providers, including vision/tool/reasoning capabilities and context/input/output limits. Task configs and both helper-config paths register these models even when no coding/helper reasoning options create entries. Existing provider SDKs, credentials, custom options, and explicit limits are preserved.
• Enable Terra through the existing ChatGPT Subscription Fast option and gateway allowlist. The Vercel catalog offers the standard Terra selection.
• Update public model documentation and focused regression tests. Existing defaults, presets, and saved selections stay unchanged.

Why this change was made

Users should be able to select the newly released Terra model through verified provider routes. Runtime definitions also prevent model lookup failures when Requesty or Azure Terra is assigned only to vision, advisor, or explore and OpenCode's refreshed catalog does not yet list it.

Impact and metadata sources

Provider metadata was verified on October 26, 2026:

Route	Upstream ID and authoritative source
OpenAI API	gpt-6.1-terra
ChatGPT Subscription	gpt-6.1-terra in Codex; Fast support, subject to plan and rollout
OpenRouter / Roomote inference	openai/gpt-6.1-terra; Roomote uses its existing namespace rebasing
Vercel AI Gateway	openai/gpt-6.1-terra
Azure OpenAI / Foundry	gpt-6.1-terra (2026-10-26); requires matching deployment name and account/region access
Requesty	azure/gpt-6.1-terra@swedencentral

OpenAI documents 1,050,000 context / 922,000 maximum input / 128,000 maximum output tokens, text/image input and text output, streaming, structured outputs, and tool calling through Responses. Reasoning supports none, low, medium, high, xhigh, and max; Roomote retains its existing effort scale. Standard pricing per million tokens: $1 input / $0.05 cache read / $1.25 30-minute cache write / $5 output. Above 272K input tokens, input/cache rates double and output is 1.5x. Vercel Standard is $1/$5 input/output; the selected Requesty route is $1.20/$6 and $0.06 cache read. Azure prices vary by deployment/region.

OpenCode Zen/Go, GitHub Copilot, and Bedrock Terra routes remain unconfirmed and are not added. Azure/Requesty UI pricing metadata may remain sparse pending upstream catalog entries; their runtime definitions no longer depend on that publication. Paid inference and account entitlements were not exercised. This change does not add a computer-use subsystem.

How it was tested

• 408 focused tests passed: 127 shared-types tests, 119 worker tests, 46 helper-runtime tests, and 116 web task-model tests.
• Regression coverage includes all three providers × vision/advisor/explore-only assignment × direct/gateway mode (18 cases), plus six vision-only helper cases across generated and operator-provided config paths. Helper cases failed before the fix due to absent model entries and pass with the shared registration.
• Six real OpenCode 1.18.30 model-listing probes passed: each exact Requesty/Azure route resolves from generated task and helper configs. Probes used isolated homes and synthetic credentials, with no inference calls.
• Shared merger tests cover preserving SDKs, provider options, per-model reasoning options, explicit limits/modalities, unrelated models, immutability, and idempotence. Catalog and ChatGPT tests confirm standard Terra routes and existing subscription Fast behavior.
• pnpm lint:fast, pnpm check-types:fast (27 packages), pnpm knip, changed-file oxfmt --check, git diff --check, and pnpm --filter @roomote/docs check passed. All pre-push gates passed with hooks enabled.
• Optional commit-time Roomote Judgement evaluation reported incomplete evidence and timed out without blocking the commit; its full evaluation remains outstanding. CI completion has not been checked.
• Browser capture was not applicable to the runtime lookup correction; this cycle retains no screenshot/video evidence.

Checklist

• The PR title follows the repo convention: [Fix], [Feat], [Improve], [Refactor], [Docs], or [Chore] followed by a user-facing description
• This PR is small and scoped to one change
• pnpm lint and pnpm check-types pass locally — repository-recommended fast equivalents and changed-file formatting passed; full formatting-inclusive aliases were not run
• I added tests or included a clear manual validation note above
• I removed secrets, tokens, private keys, and customer data from code, logs, and screenshots
• If this change should appear in the changelog, I ran pnpm changeset — no changeset added; release preparation can include this model addition in its missing-notes audit

[roomote-community]

No new code issues found. The prior runtime-registration finding is fixed. See session

Confirmed generated task config registers all three distinct-role Terra routes, and helper configs register them with both generated and existing provider configuration. Current-head CI tests, lint, type checking, knip, and docs validation passed; some build and analysis checks were still running when inspected.

• P2: Register Terra runtime model definitions for Requesty and both Azure providers — packages/types/src/model-provider-config.ts:827 (also lines 944 and 987). Fixed by shared runtime definitions applied to task and helper config builders, including role-only selections.

_{Reviewed 06b9791}

[roomote-roomote]

Fixed and pushed in 06b97918.

• Removed the separate Vercel Terra Fast selection, provider ID/map, catalog entry, documentation and test references. Standard Terra and the existing ChatGPT Subscription Fast option remain supported.
• Added shared runtime definitions for Requesty and both Azure providers so vision/advisor/explore-only selections resolve independently of upstream catalog publication or coding/helper reasoning options. Applied to task execution and both helper-config paths; existing SDKs, credentials and overrides are preserved.
• Addressed the one actionable finding in the live review, replied with evidence, and resolved its thread. PR metadata now reflects the final shipped diff and remains draft.

Validation: 408 focused tests passed (127 types, 119 worker, 46 helper, 116 web), including the 18-case task role/provider/gateway matrix and six helper-config cases. Six actual OpenCode 1.18.30 model-listing probes passed with isolated configs/synthetic credentials and no inference requests. Lint, typecheck, knip, formatting, docs checks, diff checks and pre-push gates passed.

Remaining validation limits: optional commit-time Roomote Judgement evaluation reported incomplete evidence/timeouts; full evaluation is still outstanding. Paid inference and account entitlements were not exercised, and CI completion has not been checked. No actionable review findings remain from the fetched review inventory.