Skip to content

feat(models): refresh Command Code catalog to 1.83.0 - #167

Merged
pierreraby merged 2 commits into
patlux:mainfrom
sandexzx:chore/models-catalog-1.83.0
Oct 17, 2026
Merged

pierreraby merged 2 commits into
patlux:mainfrom
sandexzx:chore/models-catalog-1.83.0

Conversation

@sandexzx

@sandexzx sandexzx commented Oct 16, 2026 •

Copy link
Copy Markdown
Contributor

Summary

Sync the static capability snapshot from command-code@1.82.4 to command-code@1.83.0.

  • Add image input, reasoning, and selectable low/medium/high efforts for meta/muse-spark-open and nvidia/nemotron-3.5-super-vl.
  • Publish reasoning for the free stealth preview stealth/cat-jam:free with low/medium/high/xhigh/max efforts and a 131,072-token output limit.
  • Retire inclusionai/ling-3.0-flash-sante:free: drop its reasoning flag, output limit, and display-price entry now that the API no longer advertises it.
  • Refresh both fixtures and the affected pricing test expectations.

Supersedes the metadata-only automation PR #166, which targeted command-code@1.82.5.

Counts move from 68 to 70 image-capable models, 77 to 79 reasoning models, and 61 to 64 models with selectable efforts.

Pricing review

No rates changed for retained models. PRICING_LAST_VERIFIED remains 2026-10-15. The only change to MODEL_COSTS is removal of the retired free Ling 3.0 Flash entry, which the API no longer advertises. The live checker reports PASS with no drift, so the reviewed snapshot now covers all 92 advertised models. Existing context tiers, DeepSeek time windows, and promotions are untouched.

Scope and limitations

  • Metadata refresh plus the retirement of one model; no transport, runtime, or command behavior changes.
  • No live authenticated generation was performed. Verification relies on the catalog/pricing checkers, the unit suites, and the isolated mock host suites.

Verification

  • npm run check:commandcode-catalog: current command-code@1.83.0, no drift.
  • npm run check:commandcode-pricing: PASS, 92 models, no pricing drift.
  • npm run test:models, npm run test:pricing: PASS.
  • npm run typecheck, npm run format:check, git diff --check: PASS.
  • Full npm test: PASS.

Synchronize the static capability snapshot with command-code@1.83.0.

Add image input, reasoning, and selectable low/medium/high efforts for
meta/muse-spark-open and nvidia/nemotron-3.5-super-vl. Publish reasoning
for the free stealth/cat-jam:free preview with
low/medium/high/xhigh/max efforts and a 131,072-token output limit.

Retire inclusionai/ling-3.0-flash-sante:free: drop its reasoning,
efforts, output limit, and display-price entry now that the API no longer
advertises it. Display prices were re-checked against the official page
and are unchanged; the reviewed snapshot now covers 92 advertised
models.
Ling 3.0 never had selectable reasoning efforts, and the retired free
entry is the only change to MODEL_COSTS. Fix both statements in the
changelog.

@pierreraby pierreraby left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

APPROVED — review at exact head 5e343b45088db353b41c8dcbdc9a33834feae0f3 (2 commits), base 3e43a7c. 7 files, +16/−11, .github/ untouched.

Verified

Catalog vs published command-code@1.83.0 (tarball SHA1 45d327d674d1ad1539bc74a140bc9045738367de): check:commandcode-model-metadata command-code@1.83.0 reports zero drift (image 70/70, reasoning 79/79, efforts 64/64, output limits 10/10). Manual bundle cross-check: meta/muse-spark-open and nvidia/nemotron-3.5-super-vl carry inputModalities:["text","image"], reasoning:!0, efforts ["low","medium","high"]; stealth/cat-jam:free carries reasoning, ["low","medium","high","xhigh","max"], output 131072; inclusionai/ling-3.0-flash-sante:free has 0 occurrences in the bundle — the retirement is justified.

Prices vs official pricing page (fetched 2026-10-17): muse-spark-open 1.25/4.25/0.15 ✓, nemotron-3.5-super-vl 0.2/0.5/0.04 ✓, cat-jam:free 0/0/0 with the 100% cat-jam-free deal "while the stealth preview lasts" ✓, ling-3.0 free entry gone ✓.

Local runs (isolated worktree at exact head, npm ci --ignore-scripts): typecheck, format:check, git diff --check all clean; full npm test exit 0 including [pi-local] PASS (pi 1.3.0) and [omp-compat] PASS.

RED-on-base proved: PR tests + fixtures copied onto 3e43a7c make test-pricing fail (snapshot/coverage mismatch); green at head. CI at exact head: 13 success + 2 skipped (sync, Update PR comment).

Internal consistency: every added/removed catalog entry has its counterpart (modalities ↔ reasoning ↔ efforts ↔ output limits ↔ pricing ↔ fixtures ↔ tests), no orphans; ling-3.0 removed everywhere except the changelog; deepEqual(MODEL_COSTS, expected) pins prices + the removal, pricedIds == catalogIds pins coverage, and the freeModels check pins free status.

Non-blocking notes

  1. P2 — supersedes #166: verified — #167's catalog is strictly #166's (only version strings 1.82.5→1.83.0 differ). Merging #167 alone suffices; #166 can be closed as superseded with credit to Ömer's review.
  2. P2 — upstream CLI is already 1.84.0 (published 16 Oct 18:29Z), but comparing the 1.83.0 and 1.84.0 bundles shows identical models.md and identical model entries — 1.84.0 brings no metadata change, so this PR is not stale on substance; the sync bot will just bump the version string later.
  3. P2 — reflection/beam:free is advertised by the live API (93 models) but absent from the fixture (92, fetched 16 Oct 17:58) and from both CLI bundles — server-side drift newer than this snapshot, out of scope; check:commandcode-pricing flags it as missing-local-price (informational REVIEW REQUIRED, expected).
  4. P3 — assert.match(fixture.fetchedAt, /^2026-10-16T/) pins the fetch day; must be bumped on every refresh (existing practice; ^2026-10- would be less brittle).
  5. P3 — PRICING_LAST_VERIFIED remains 2026-10-15 although the ling retirement was re-verified against the live page on 17 Oct; consistent with the fixture check, just under-dated.

Method note

The usual second fresh-context reviewer pass is deferred: subagent launches currently fail on a session-directory resolution error (unrelated to this PR). All checks above were run directly at the exact head. No billed probes were used (public pricing page + npm packages only).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants