Skip to content

chore(models): refresh catalog to 1.88.0 and price Kimi K3.1 Pro - #173

Merged
pierreraby merged 1 commit into
mainfrom
chore/models-catalog-1.88.0
Oct 23, 2026
Merged

pierreraby merged 1 commit into
mainfrom
chore/models-catalog-1.88.0

Conversation

@pierreraby

Copy link
Copy Markdown
Collaborator

Summary

Refreshes the catalog after the 2026-10-23 upstream movement and closes the gap where a new paid model displayed a zero cost.

  • CLI snapshot → command-code@1.88.0 (via npm run sync:commandcode-catalog). The only metadata delta is that upstream drops image input for xiaomi/mimo-v2.5; reasoning flags, selectable efforts and output limits are unchanged (72 → 71 image models).
  • New pricing: moonshotai/Kimi-K3.1-Pro at $2 input / $10 output / $0.2 cache read per million tokens, cache write free — read from the official pricing page on 2026-10-23, not from the API. Without this entry the model silently displayed zero cost. The CLI does not publish capability metadata for it yet (its reference lists 92 ids), so it keeps the conservative text-only, non-reasoning defaults with a 65,536-token output limit until upstream catches up.
  • Retirements: xiaomi/mimo-v2.5 and xiaomi/mimo-v2.5-pro are gone from both the live API and the pricing page (replaced by the v2.6 family), so their display prices are removed. The API now advertises 93 models.
  • Fixtures refreshed (93 ids, 93 prices, fetchedAt 2026-10-23T20:45:35Z), PRICING_LAST_VERIFIED → 2026-10-23.

Verification

  • check:commandcode-catalog: Current against command-code@1.88.0.
  • check:commandcode-pricing: PASS, no drift, no missing local price.
  • npm test: PASS (full suite, including [pi-local] and [omp-compat]).
  • npm run typecheck, npm run format:check, git diff --check: clean.
  • Test expectations updated to the verified upstream values (Kimi-K3.1-Pro spot-check replaces the retired MiMo assertion); nothing weakened.

Notes

@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 d4544b6 with PR head eabc5ac on the same GitHub-hosted darwin runner. Lower values are better.

Metric Base PR PR − Base Change
Stable RSS 120.0 MiB ± 0.5 MiB 120.6 MiB ± 0.6 MiB +0.4 MiB ± 1.0 MiB +0.4%
Physical footprint 66.2 MiB ± 1.3 MiB 67.5 MiB ± 0.1 MiB +1.4 MiB ± 1.1 MiB +2.1%
Physical peak 77.4 MiB ± 1.4 MiB 78.0 MiB ± 0.2 MiB +0.8 MiB ± 1.4 MiB +1.0%

Interpretation: Differences larger than their measured MAD: Physical footprint +1.4 MiB ± 1.1 MiB.

Extension overhead above pi baseline

Metric Base overhead PR overhead Difference
Stable RSS +6.7 MiB ± 0.9 MiB +7.8 MiB ± 0.5 MiB +0.4 MiB ± 1.0 MiB
Physical footprint +4.0 MiB ± 1.4 MiB +5.4 MiB ± 0.7 MiB +1.4 MiB ± 1.1 MiB
Physical peak +5.9 MiB ± 2.0 MiB +6.5 MiB ± 1.4 MiB +0.8 MiB ± 1.4 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 1.4.0, 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
pierreraby merged commit ba524b2 into main Oct 23, 2026
15 checks passed
@pierreraby
pierreraby deleted the chore/models-catalog-1.88.0 branch October 23, 2026 20:50
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