[sandexzx]

Summary

Continue the original catalog/pricing contribution on this existing branch, preserving its commits.

• Regenerate static capabilities from command-code@1.82.4: Gemini 4 Argon, Muse Wonder, Qwen 3.8 Max Prime, and paid Ling 3.1 Flash (replacing its retired free ID).
• Fix the reproduced catalog-sync parser failure: CLI 1.82.4 removed the global text-only Set and added behaviorProfile before inputModalities. Read each model's explicit modalities, reject missing/unsupported metadata, and do not advertise the CLI-only file modality. Regression tests failed before the fix and pass afterward.
• Refresh the live API fixture and display prices for all 93 advertised models as of 2026-10-15, including the newly advertised Muse Spark Open, Nemotron 3.5 Super VL, and free Cat Jam preview.
• Update model/pricing regression assertions and changelog.

Pricing review

Reviewed against https://commandcode.ai/docs/resources/pricing-limits on 2026-10-15. USD per million tokens, input / output / cache read / cache write:

Model	Rates
Gemini 4 Argon	2 / 10 / 0.10 / 0
Muse Wonder	4 / 20 / 0.20 / 0
Qwen 3.8 Max Prime	4 / 12 / 0.50 / 5
Ling 3.1 Flash	0.30 / 0.90 / 0.06 / 0
Nemotron 3.5 Super VL	0.20 / 0.50 / 0.04 / 0
Muse Spark Open	1.25 / 4.25 / 0.15 / 0
Cat Jam preview	0 / 0 / 0 / 0

Cat Jam is free only while the stealth preview lasts; no fixed expiry is published. Existing context tiers, DeepSeek time windows, and promotions remain unchanged. Boosted subscription allowances do not change per-token prices. The existing reviewed expired Qwen badge remains visible in the checker report.

Scope and limitations

• Supersedes the metadata-only automation PR chore(models): sync Command Code catalog #151.
• The API is newer than CLI 1.82.4. Nemotron 3.5 Super VL, Cat Jam, and Muse Spark Open are discoverable and priced, but no unverified static image/reasoning/effort metadata is invented. Existing host-resolved metadata and conservative defaults apply.
• No live authenticated generation was performed. Host integration uses isolated loopback mock APIs and temporary credentials/configuration.

Verification

• npm run check:commandcode-catalog: current 1.82.4; 68 image, 77 reasoning, 61 effort, 10 output-limit entries match upstream.
• npm run check:commandcode-pricing: PASS, 93 models, no pricing drift.
• Full npm test with Pi 1.3.0 and OMP 18.11.1 required: PASS, including both real-host mock suites.
• Typecheck, formatting, and git diff --check: PASS.
• Exact updated head must pass fresh GitHub CI before merge; the previous approval predates these follow-ups.

[pierreraby]

Verified exact head 7402cf930f9710d20f8682d693dc5d4a87861a6a (fork sandexzx:chore/models-catalog-1.82.1, base b61093d) — catalog refresh to command-code@1.82.1 with reviewed pricing.

Independent checks on a clean detached checkout of this exact head:

• Pricing verified against the live official page (commandcode.ai/docs/resources/pricing-limits): google/gemini-4-argon $2/$10/$0.10 and meta/muse-wonder-1.0 $4/$20/$0.20 per 1M in/out/cache-read, no cache-write rate — exact match in both src/pricing.ts and fixtures.
• npm run typecheck PASS; full npm run test:unit green (17 files, 0 failures, exit 0); npm run test:pricing 10/10.
• npm run check:commandcode-catalog — metadata current (1.82.1; 67/67 image, 76/76 reasoning, 60/60 efforts, 9/9 output limits); npm run check:commandcode-pricing — PASS, no drift.
• prettier --check and git diff --check clean.
• Scope: exactly 7 files (CHANGELOG, README, catalog, pricing, 2 fixtures, test-pricing); .github/ byte-identical to base; upstream checks on this head all green; mergeable_state: clean.
• This supersedes the metadata portion of automation #151 (1.81.0, no pricing review) and adds the pricing review the bot does not perform.

One P2 note: the offline pricing-page HTML fixture (~2 weeks stale) does not exercise the two new rows through the parser; live coverage rests on check:commandcode-pricing — consider refreshing it in a future sync.

Approving. Leaving merge timing and #151 disposition to the maintainers.