Skip to content

chore(models): refresh November 17 catalog and Flint pricing - #223

Merged
pierreraby merged 1 commit into
mainfrom
chore/catalog-1.89.2
Nov 17, 2026
Merged

pierreraby merged 1 commit into
mainfrom
chore/catalog-1.89.2

Conversation

@pierreraby

Copy link
Copy Markdown
Collaborator

Summary

Follow up #220 with the upstream changes published after its snapshot:

  • Refresh generated metadata to command-code@1.89.2 (80 image / 90 reasoning / 74 effort entries / 10 output limits).
  • Advertise image + reasoning + selectable efforts for Qwen/Qwen4-Max (off, low, high, xhigh) and tencent/hy4-flash (off, high).
  • Add text-only reasoning for recursive/flint-v1.0:free, without inventing selectable efforts or a model-specific output limit.
  • Add Flint's reviewed zero display prices while its undated free tier lasts; no fixed expiry is published, so no dated temporary-pricing entry is invented.
  • Refresh the public API/pricing snapshots to 101 models, verified 2026-11-17, and correct the Unreleased wording following chore(models): price the November 16 catalog additions #220.

All three use the existing 65,536-token default output cap, bounded by context, and the existing endpoint-selection rule prefers Responses when advertised. No runtime/routing rule, workflow, existing rate, or model-specific output limit changes.

Sources

Validation

Fresh pre-publication checks:

  • Live catalog check: Current at 1.89.2; live pricing check: PASS, all 101 models, no drift.
  • Full npm test with Pi and OMP required: PASS against isolated mocks, including protocol and endpoint-routing suites.
  • Unit typecheck, entrypoint typecheck against installed Pi 1.5.2, formatting and git diff --check: PASS.
  • Dedicated model-list regressions for the three IDs cover contexts, Responses selection, reasoning/modalities/efforts and output cap; these are not claimed as dedicated end-to-end inference tests.
  • Explicit zero-price/undated-term regression for Flint and page-ID candidate assertions for all three IDs.

Scope / limitations

Workflow action pinning is a separate independent PR, #222; the workflows on this branch remain identical to main. Existing auxiliary workflow alerts are not claimed resolved here. Inferred-project LSP reports omit host declarations; the dedicated entrypoint compiler check against the real installed host passes. No global all-scanners-clean verdict is claimed.

No account login, billable inference, package installation or production config update was performed. This PR is for review/CI only; no merge or release is authorized.

Sync command-code 1.89.2 capabilities and review the 101-model API snapshot. Add undated free-tier pricing for Flint and explicit model/routing and pricing-page ID regressions. Correct the Unreleased description following #220.
@github-actions

Copy link
Copy Markdown
Contributor

Dependency Review

✅ No vulnerabilities or license issues or OpenSSF Scorecard issues found.

Scanned Files

None

@github-actions

Copy link
Copy Markdown
Contributor

Memory benchmark

Compared base 7d74f45 with PR head 773c9e3 on the same GitHub-hosted darwin runner. Lower values are better.

Metric Base PR PR − Base Change
Stable RSS 105.9 MiB ± 0.1 MiB 107.1 MiB ± 0.8 MiB +0.8 MiB ± 0.7 MiB +0.7%
Physical footprint 60.4 MiB ± 0.4 MiB 60.9 MiB ± 0.7 MiB +0.5 MiB ± 0.8 MiB +0.8%
Physical peak 71.4 MiB ± 0.5 MiB 72.2 MiB ± 1.6 MiB +1.3 MiB ± 0.7 MiB +1.8%

Interpretation: Differences larger than their measured MAD: Stable RSS +0.8 MiB ± 0.7 MiB, Physical peak +1.3 MiB ± 0.7 MiB.

Extension overhead above pi baseline

Metric Base overhead PR overhead Difference
Stable RSS +3.7 MiB ± 0.9 MiB +5.2 MiB ± 0.6 MiB +0.8 MiB ± 0.7 MiB
Physical footprint +2.2 MiB ± 0.5 MiB +2.5 MiB ± 0.8 MiB +0.5 MiB ± 0.8 MiB
Physical peak +1.1 MiB ± 0.2 MiB +3.1 MiB ± 1.3 MiB +1.3 MiB ± 0.7 MiB

Values are medians of 6 alternating, paired runs. The value after ± is the median absolute deviation (MAD). Each process was sampled 12 times after a 2500 ms warm-up.

Environment: pi 0.84.4, Bun 1.4.4, darwin arm64. RSS comes from ps; physical footprint and peak come from macOS footprint.

This is a comparative signal, not a pass/fail threshold. GitHub-hosted runner noise can affect absolute values.

@pierreraby

Copy link
Copy Markdown
Collaborator Author

Final maintainer review at exact head 773c9e374371e9b43a4bba177d86afb3b57b8aaf: the nine-file diff matches the reviewed 1.89.2/101-model refresh, with Flint's undated zero prices and explicit model/page-ID regressions. No runtime or workflow change is included. Fresh live catalog/pricing checks, model/pricing/checker tests and the real-host entrypoint typecheck pass; the full Pi/OMP mock suite passed before publication and all PR checks remain green.

The operator has now explicitly authorized the reviewed integration plan: merge this catalog refresh first, bring #222 up to date, then merge #222 only after its new exact-head CI passes. This is not an auto-merge based on elapsed time, and does not authorize a release or production update.

Existing inferred-project LSP and unchanged-workflow auxiliary diagnostics remain documented; this is not a claim that every scanner/worktree is clean. No billable inference was performed.

@pierreraby
pierreraby merged commit c995c01 into main Nov 17, 2026
15 checks passed
@pierreraby
pierreraby deleted the chore/catalog-1.89.2 branch November 17, 2026 09:18
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.

1 participant