lidge-jun · Sep 29, 2026 · Sep 29, 2026 · Sep 29, 2026 · Sep 29, 2026 · Sep 29, 2026
@@ -0,0 +1,21 @@
+# 260930 GPT-6.1 Sol rollout, TokenLab protocol follow-up, release
+
+Status: open. Loop session `01a0ef42-da06-7200-8394-aa35ebbe4ba9`, branch `codex/gpt-6-1-sol-rollout` from `origin/dev` `b78bfb8f00`.
+
+OpenAI released GPT-6.1 Sol on 2026-09-29 as the successor to GPT-6 Sol. Only Sol moved to 6.1; Astra and Luna stay on GPT-6. This unit adds the model everywhere GPT-6 Sol is served, moves every place where GPT-6 Sol is the *default* to GPT-6.1 Sol (the same move #5640 made from GPT-5.6 to GPT-6), folds in TokenLab's protocol request from mail 546, and ships a release.
+
+| Doc | Work-phase | Content |
+|---|---|---|
+| 010_research_digest.md | wp1 | Sourced facts for GPT-6.1 Sol and the TokenLab contract |
+| 020_catalog_surfaces.md | wp2 | Diff-level list of provider, catalog, pricing and docs rows |
+| 030_default_swap.md | wp2 | Defaults moving from gpt-6-sol to gpt-6.1-sol, roster migration v3 |
+| 040_tokenlab_protocols.md | wp3 | Per-model wire routing; TokenLab JEV decision backend deferred to its own unit |
+| 050_release.md | wp4 | PR, CI, merge, preview/main promotion, release.yml, npm verification |
+
+## Audit record (wp1)
+
+Read-only reviewer on gpt-6-sol (high), 2026-09-30:
+
+- Round 1 FAIL, four blockers. Three concerned the JEV backend (combo dispatch passes no combo settings, combo normalization/persistence and the GUI editor would drop new fields, credential/URL/outbound guard must switch together). One named roster tests that enumerate models by value.
+- Fold: the roster tests are listed in 020; the JEV backend is deferred to its own unit (040 records why and sketches it).
+- Round 2 PASS. Non-blocking notes kept for B: the v3 roster migration must run after the v2 step and touch only bare `gpt-6-sol`; TokenLab's anthropic wire sends `x-api-key` to `/v1/messages`, which TokenLab accepts; `kiro-adapter.test.ts` (2047/2050) and `codex-catalog.test.ts` (7974/7985) take in-place edits only.
@@ -0,0 +1,53 @@
+# 010 Research digest
+
+Observed 2026-09-30 KST. Full report with every URL: Aside run, `/Users/jun/.aside/u/0/artifacts/gpt61sol-research.md` (kept outside the repo). Live probes re-run from this checkout are marked (live).
+
+## GPT-6.1 Sol, official
+
+| Fact | Value | Source |
+|---|---|---|
+| API id | `gpt-6.1-sol`, no dated snapshot | developers.openai.com/api/docs/models/gpt-6.1-sol |
+| Released | 2026-09-29 (API, ChatGPT Work, Codex) | openai.com/index/introducing-gpt-6-1-sol/ |
+| API context / max input / max output | 1,050,000 / 922,000 / 128,000 | model page |
+| API efforts | low, medium (default), high, xhigh, max | model page |
+| Price per 1M | $2 input, $0.10 cached, $2.50 cache write, $10 output | pricing.md |
+| Long context (>272K prompt) | $4 / $0.20 / $5 / $15 | pricing.md |
+| Fast | 2x standard | pricing.md |
+| Modalities | text + image in, text out | model page |
+| Codex row (live, openai/codex models.json, #49318) | priority 1 (catalog default), efforts low..ultra, default effort low, context 272,000 / 872,000, Fast tier "2x speed, increased usage", minimal client 0.153.0 | codex-rs/models-manager/models.json |
+| gpt-6-sol | retained: visibility list, priority 3, no upgrade, not deprecated | same file; deprecations page |
+| GPT-6.1 Luna / Astra | do not exist (doc pages 404, absent from pricing, models index and Codex catalog) | model pages |
+
+Only cached input changed versus GPT-6 Sol ($0.20 -> $0.10).
+
+## Third-party listing (drives 020)
+
+| Provider in this repo | Status | Id |
+|---|---|---|
+| OpenRouter | listed | `openai/gpt-6.1-sol` (1,050,000 ctx, 128,000 out) |
+| Vercel AI Gateway | listed | `openai/gpt-6.1-sol` |
+| GitHub Copilot | listed, GA rollout (github.blog 2026-09-29) | `gpt-6.1-sol` |
+| Kilo | listed (models.dev) | `openai/gpt-6.1-sol` |
+| OpenCode Zen | listed (live /zen/v1/models) | `gpt-6.1-sol` |
+| Devin | listed (devin.ai blog); id not published | preemptive `gpt-6-1-sol` following its `gpt-5-6-sol` spelling |
+| TokenLab | listed (live /v1/models/gpt-6.1-sol, Chat + Responses) | `gpt-6.1-sol` (live discovery) |
+| Amazon Bedrock | not in models.dev; openai/codex source references `openai.gpt-6.1-sol` | preemptive |
+| Cloudflare AI Gateway | not in catalog; OpenAI passthrough | preemptive |
+| Kiro, CodeBuddy | not listed / unreadable; both carry preemptive gpt-6-sol rows today | preemptive, same policy as the Sonnet 5.5 Kiro rows |
+| ZenMux | page explicitly "not a registered ZenMux page" | not added (live discovery picks it up) |
+| DigitalOcean, Scaleway | not listed; DO naming for 6.1 unknown | not added |
+
+## TokenLab (mail 546, 2026-09-30; live contract re-checked)
+
+`GET https://api.tokenlab.sh/v1/models/{id}` returns `tokenlab.accepted_request_formats` (live):
+
+| Models | Formats |
+|---|---|
+| gpt-6-astra, gpt-6-sol, gpt-6-luna, gpt-6.1-sol | chat, responses |
+| claude-opus-5, claude-opus-5-5, claude-sonnet-5, claude-sonnet-5-5, claude-fable-5, claude-fable-5-1 | chat, anthropic_messages |
+| deepseek-v4.1-flash, deepseek-v4-pro, kimi-k3, glm-5.3 | chat, responses, anthropic_messages |
+| grok-4.7 | chat, responses |
+| gemini-3.8-flash | chat, gemini_generate_content |
+
+Endpoints from docs.tokenlab.sh/llms.txt: `/v1/chat/completions`, `/v1/responses`, `/v1/messages` (x-api-key or Bearer), `/v1/systemone`. System One takes `{model, state, questions}` with Bearer auth, current model `jev-1.13` — the same body shape `src/combos/jev.ts` already sends to TypeSafe. Vincent asks to keep the user's API-key delivery policy default (no forced `X-TokenLab-Delivery-Policy`).
+
@@ -0,0 +1,30 @@
+# 020 Catalog surfaces for gpt-6.1-sol (wp2)
+
+Rule: add a row wherever a gpt-6-sol row exists and 010 says listed or preemptive; comment preemptive rows `260930 preemptive`. No gpt-6.1-luna / gpt-6.1-astra anywhere.
+
+| File | Change |
+|---|---|
+| scripts/model-metadata.source.json | Clone the gpt-6-sol row as gpt-6.1-sol in openai, openai-codex, github-copilot, kilo, openrouter, vercel-ai-gateway, opencode-zen, amazon-bedrock (preemptive), cloudflare-ai-gateway (preemptive); cache_read 0.10 (0.20 on opencode-zen per its published figure); release_date 2026-09-29; knowledge 2026-04-30. ZenMux not added. Then `bun run generate:model-metadata`. |
+| src/usage/expected-prices.ts | `GPT61_SOL` cost4 (2 / 10 / 0.10 / 2.50) with long-context tier; rows for openai-apikey (verified), openai (verified-derived), devin, devin-cli (verified-derived); pricing-list id. |
+| src/providers/registry/model-seeds.ts | `OPENAI_GPT6_MODELS` gains gpt-6.1-sol (flows to openai-apikey 1,050,000 / 922,000 / 128,000 / low..max and OpenRouter models); OpenRouter context map adds openai/gpt-6.1-sol. |
+| src/providers/registry/entries-core.ts | OpenRouter `modelSupportsServiceTier` adds openai/gpt-6.1-sol; BizRouter seed adds openai/gpt-6.1-sol; Devin seed adds gpt-6-1-sol (preemptive). |
+| src/providers/registry/entries-extended.ts | Copilot models + `modelWireDefaults` gpt-6.1-sol -> openai-responses. |
+| src/providers/codebuddy-models.ts, src/providers/kiro-models.ts, src/adapters/kiro/reasoning.ts, src/adapters/devin/live-models.ts | Mirror the gpt-6-sol rows (preemptive). |
+| src/codex/catalog/native-models.ts | `NATIVE_GPT61_SOL_MODEL = "gpt-6.1-sol"` in built-in list, SELF_DESCRIBED set, drain-sentinel set; configured-native template moves to gpt-6.1-sol. |
+| src/codex/catalog/metadata.ts | DOCUMENTED_NATIVE_OPENAI_ADDITIONS and context map (NATIVE_GPT6_CONTEXT). |
+| src/codex/catalog/effort.ts | Ladder entry low..ultra like Sol. |
+| src/codex/data/roster-pinned-models.json | Append the upstream gpt-6.1-sol row verbatim from openai/codex models.json (codex-rs bundle, #49318); pinned-models.ts comment updated. |
+| src/codex/model-entitlements.ts | Comment: 6.1 Sol ungated like Sol. |
+| docs-site providers.md, structure/catalog.md, structure/providers/openai-accounts.md | Document the row. |
+
+Tests that enumerate the roster by value and must gain gpt-6.1-sol (audit A1 blocker 4):
+
+- tests/codex-integration/native-model-toggle.test.ts:95 (built-in native roster) and :127 (flagship list)
+- tests/providers/provider-registry-parity.test.ts:314 (exact openai-apikey model list)
+- tests/codex-integration/codex-catalog.test.ts:6842 (catalog roster; file is 7974/7985 lines, so edit in place without adding lines)
+- tests/providers/github-copilot/github-copilot-wire-defaults.test.ts:33, tests/service/service-tier-capability.test.ts:73 (OpenRouter tier map), tests/routing/subagent-model-fallback.test.ts:276
+- tests/providers/kiro/kiro-adapter.test.ts:1744 is at 2047/2050: edit the existing list in place only.
+- tests/codex-integration/configured-native-models.test.ts:93-102: a configured native must now borrow gpt-6.1-sol's pinned capabilities and hash (wp2 audit NEAR-PASS item). catalog-routed-comp-hash.test.ts:75 keeps its gpt-6-sol forward-alias expectation.
+- BizRouter is not added: 010 has no evidence it lists gpt-6.1-sol, and live discovery picks it up once it does.
+
+New focused test `tests/codex-integration/gpt61-sol-rows.test.ts` (registered in scripts/test-layout/layout.json and tests/fixtures/test-layout-expected.json) asserting: native row ladder low..ultra, default effort low, 272,000/872,000 context; openai-apikey 1,050,000/922,000/128,000 and low..max; Copilot Responses wire; expected price 2/10/0.10/2.50; no gpt-6.1-luna/astra id in any registry entry.
@@ -0,0 +1,14 @@
+# 030 Default swap gpt-6-sol -> gpt-6.1-sol (wp2)
+
+Precedent: #5640 (`8adda594fd`) set the subagent default to the GPT-6 trio and upgraded stored rosters from 5.6 once, by version.
+
+| Surface | Change |
+|---|---|
+| src/config/subagent-models.ts | `DEFAULT_SUBAGENT_MODELS = [astra, gpt-6.1-sol, luna]`; `SUBAGENT_MODELS_VERSION = 3`; the v3 step replaces bare `gpt-6-sol` with `gpt-6.1-sol` in place (deduped), once. A later deliberate re-pick of gpt-6-sol stays. v0/v1 installs chain through v2 then v3. |
+| scripts/ci/docker-smoke.ts | subagentModels default trio. |
+| Codex configured-native template | borrows gpt-6.1-sol's row (020). |
+| docs-site agents.md (10 locales), guides/claude-code.md (10 locales), structure/subagents.md | Name the new default. |
+| Tests | subagent-roster-migration (v2 -> v3, idempotence, re-pick retained, routed ids untouched), claude picker/intercept/management expectations that read the default trio, server startup reconcile. |
+
+Out of scope: removing gpt-6-sol anywhere; it stays listed and selectable.
+
@@ -0,0 +1,20 @@
+# 040 TokenLab per-model protocols and JEV backend (wp3)
+
+## Wires
+
+Keep the preset's provider-wide adapter `openai-chat` (the released, verified path). Add registry defaults so each model rides its declared native wire:
+
+- `modelWireDefaults` -> `{ wire: "openai-responses", inbound: ["responses"] }` for gpt-6-astra, gpt-6-sol, gpt-6-luna, gpt-6.1-sol, grok-4.7, deepseek-v4.1-flash, deepseek-v4-pro, kimi-k3, glm-5.3. Scoped to Responses inbound (Codex) like the Alibaba Token Plan precedent (tests/providers/alibaba-token-plan-wire-defaults.test.ts): a Chat or Anthropic client keeps the verified Chat wire with no translation hop. User-overridable: an explicit `modelAdapters` entry of `openai-chat` wins.
+- Claude ids -> Anthropic Messages through the existing endpoint-bound prefix pin (`WIRE_ADAPTER_PIN_PREFIXES` in src/types/wire.ts, the Command Code mechanism): `tokenlab: { endpoint: "https://api.tokenlab.sh/v1", prefixes: { "claude-": "anthropic" } }`. The anthropic adapter already normalizes `/v1` to `/v1/messages`. The pin is bound to the canonical endpoint, so a retargeted TokenLab row is untouched. Every live `claude-*` TokenLab id is checked to declare anthropic_messages before the prefix is used.
+- Everything else, including gemini-3.8-flash, stays on Chat. Gemini native is deferred until tested, as Vincent proposed.
+- No delivery-policy header is added: the user's API-key default stays authoritative.
+
+Rejected: a provider-wide switch to Responses (sends Claude/Gemini to an endpoint they do not accept); separate provider entries (duplicates models by default); widening `MODEL_ADAPTER_OVERRIDE_ALLOWED` to anthropic (needs the #404 credential threat model; the prefix pin already exists and is endpoint-bound).
+
+Tests: resolveWireProtocolOverride / resolved policy for each class on the canonical endpoint, user modelAdapters override back to Chat for a Responses default, retargeted base URL keeps Chat, anthropic URL resolves to https://api.tokenlab.sh/v1/messages.
+
+## JEV decision backend — deferred to its own unit
+
+Decision after audit A1 (blockers 1-3): not in this release. A per-combo backend has to travel through `src/server/responses/core-combo.ts:536` (the call passes no combo settings), combo normalization and persistence (`src/combos/types.ts:379`, `src/server/management/combo-routes.ts:231`), and the GUI combo editor round trip (`gui/src/combo-workspace-data.ts:266`, `:460`), and credential, fixed URL and canonical outbound guard (`src/combos/jev.ts:568`, `:599`) must switch together. It also sends conversation-derived decision state to a new third party, which deserves its own review and a GUI screenshot. That does not meet this unit's "fits cleanly" bar.
+
+Follow-up unit sketch: combo fields `jevBackend: "typesafe" | "tokenlab"` (default typesafe) and `jevModel` (`jev-*`), fixed endpoint per backend (`https://api.typesafe.ai/v1/systemone`, `https://api.tokenlab.sh/v1/systemone`, body `{model, state, questions}` confirmed identical in docs.tokenlab.sh/api-reference/systemone/create-decision), credentials never shared between backends, allowlist/timeout/cancellation/fail-open unchanged.
@@ -0,0 +1,39 @@
+# 050 Release (wp4)
+
+1. Isolated verification in a /private/tmp checkout of the exact head with fresh OPENCODEX_HOME / CODEX_HOME: typecheck, focused tests, test:changed, structure:check, privacy:scan, skill:surface:check, file-size ratchet.
+2. One PR to dev from codex/gpt-6-1-sol-rollout with the template; wait for exact-head required CI; maintainer merge under MAINTAINERS.md dev policy.
+3. Release train per scripts/release.ts and the 2.70.0 precedent: dev version pre-move PR, preview promotion PR, main promotion PR; push-event Cross-platform CI + Service lifecycle green on each promotion SHA; release.yml dispatched with expected-sha for preview then stable.
+4. Verify GitHub release assets, npm `latest` / `preview` dist-tags, latest.json.
+5. Close the unit: move to devlog/_fin with an outcome doc.
+
+## 2.73.0 concrete steps (revalidated at wp4 P, 2026-09-30)
+
+Scope, on the owner's request to bundle the RT6 stabilization chat: 2.73.0 ships everything on `dev`
+since v2.72.0 — the RT6 train (13 PRs through #6203, recorded in `devlog/_plan/260930_release_train_6/`
+by that chat), #6261 (`540af24384`, Windows keyring test portability), and this PR. That chat was asked
+not to run the release train itself. `dev` already reads 2.73.0 (#6243).
+
+1. PR from `codex/gpt-6-1-sol-rollout` (rebased on `540af24384`). Exact-head PR CI plus a
+   `lane=all` Cross-platform CI dispatch on the head (the pull_request event skips windows 1-9).
+   Before merge, an explicit read-only security review of the TokenLab wire change
+   (MAINTAINERS.md: credential handling): where the API key is sent, which header, endpoint binding
+   of the Claude pin, no new logging; verdict recorded on the PR.
+   `scripts/ci/assert-mergeable-review.sh --maintainer-integration <pr>`, decision comment,
+   `gh pr merge --admin --squash --match-head-commit <head>` -> C. Assert `git show C:package.json` = 2.73.0.
+2. Pre-move: `gh workflow run dev-version-bump.yml --ref main -f intended-version=2.73.0 -f mode=pre-move`;
+   merge its PR after CI -> dev 2.74.0.
+3. Preview: branch `codex/promote-preview-2.73.0` from C, `git merge -s ours origin/preview`,
+   `bun scripts/release-version-sources.ts sync 2.73.0-preview.20260930`; diff vs C = four version sources.
+   PR to preview, `--admin --merge --match-head-commit` after its PR checks. The maintainer-integration
+   exception covers only `dev`; promotion merges run on the owner's explicit release authorization
+   for this unit (2026-09-30, "exec and release" / "배포해줘"), as 2.70.0-2.72.0 did, and each promotion
+   PR records that authorization.
+4. Main: branch `codex/promote-main-2.73.0` from C, `git merge -s ours origin/main`, `git diff --quiet C HEAD`.
+   PR to main, same merge.
+5. Push-event Cross-platform CI + Service lifecycle success on each promotion SHA.
+6. `gh workflow run release.yml --ref preview -f version=2.73.0-preview.20260930 -f tag=preview -f dry-run=false -f expected-sha=<preview>`,
+   then `gh workflow run release.yml --ref main -f version=2.73.0 -f tag=latest -f dry-run=false -f expected-sha=<main>`
+   (release.yml defaults to a dry run and rejects other refs). Never republish; resume with
+   `resume-after-npm-publish=true` after an npm-acknowledged failure.
+7. Verify npm dist-tags, gitHead, GitHub releases (stable not prerelease, asset count as v2.72.0), latest.json.
+8. Record PR moving this unit to `devlog/_fin/`.
@@ -190,7 +190,7 @@ du sélecteur à une route opencodex :
 
 ```bash
 ocx claude desktop bind claude-sonnet-4-6 xai/grok-4.7
-ocx claude desktop bind claude-opus-4-6 native/gpt-6-sol
+ocx claude desktop bind claude-opus-4-6 native/gpt-6.1-sol
 ocx claude desktop unbind claude-opus-4-6
 ```
 

@@ -11,7 +11,7 @@ Les paramètres des agents déterminent la surface de collaboration Codex annonc
 | --- | --- | --- | --- |
 | `multiAgentMode?` | `"v1" \| "default" \| "v2"` | `"default"` | `v1` marque tous les modèles du catalogue comme compatibles v1 ; `v2` les marque tous comme compatibles v2. `default` rétablit les choix imposés en amont (Sol/Terra en v2, Luna en v1) et suit sinon l’indicateur natif `multi_agent_v2`. S’applique aux nouvelles sessions. |
 | `keepNativeChatGptOnV1?` | `boolean` | `false` | Lorsque `multiAgentMode` vaut `"v2"`, marque les lignes natives ChatGPT (Sol/Terra et les autres modèles du backend ChatGPT) comme v1. Les parents routés restent en v2. Utilisez cette option pour qu'un parent ChatGPT puisse encore lancer Grok ou Claude — les tâches enfants v2 natives sont chiffrées par le service en amont ([#92](https://github.com/lidge-jun/opencodex/issues/92)). Ignoré en `v1` et `default`. |
-| `subagentModels?` | `string[]` | `gpt-6-astra`, `gpt-6-sol`, `gpt-6-luna` | Jusqu’à cinq identifiants de modèles natifs non qualifiés, qualifiés par un compte sous la forme `<selector>/<native-openai-model>`, ou routés sous la forme `provider/model`, affichés en tête du sélecteur de sous-agents. Le tableau de bord ne propose que les identifiants natifs non qualifiés et les identifiants routés ; lors de l’enregistrement, il omet les choix exacts qualifiés par un compte. Pour les définir, utilisez `ocx agent subagents set` ou modifiez la configuration. Après la [migration unique vers Astra](/reference/configuration/agents/#astra-roster-upgrade), une liste explicitement vide est conservée. |
+| `subagentModels?` | `string[]` | `gpt-6-astra`, `gpt-6.1-sol`, `gpt-6-luna` | Jusqu’à cinq identifiants de modèles natifs non qualifiés, qualifiés par un compte sous la forme `<selector>/<native-openai-model>`, ou routés sous la forme `provider/model`, affichés en tête du sélecteur de sous-agents. Le tableau de bord ne propose que les identifiants natifs non qualifiés et les identifiants routés ; lors de l’enregistrement, il omet les choix exacts qualifiés par un compte. Pour les définir, utilisez `ocx agent subagents set` ou modifiez la configuration. Après la [migration unique vers Astra](/reference/configuration/agents/#astra-roster-upgrade), une liste explicitement vide est conservée. |
 | `injectionModel?` | `string` | — | Modèle de sous-agent natif ou routé privilégié dans les consignes de délégation v2 produites par le proxy. |
 | `injectionEffort?` | `string` | — | Niveau d’effort privilégié (de `low` à `ultra`), pertinent uniquement avec `injectionModel`. |
 | `injectionPrompt?` | `string` | — | Remplace le corps des consignes v2 intégrées. Accepte `{{model}}`, `{{effort}}`, `{{roster}}` et `{{fallback}}`. La présence d’un `injectionModel` suffit pour produire le prompt personnalisé. |