Repository navigation
feat(models): sync the hand-written catalog to the live gateway (2026-10-20) - #485
Merged
Merged
Conversation
…-10-20) Six chat models blockrun.ai/api/v1/models serves that ClawRouter did not carry, so pins skipped the balance pre-check and maxCostPerRun: openai/gpt-6.1-sol, qwen/qwen3.8-max, tencent/hy4-preview, mistral/mistral-large-4, poolside/laguna-s-2.1 and free/muse-glimmer-30b (nvidia/muse-glimmer-30b upstream). Registry, picker, explicit-pin aliases, README and SKILL.md. Every rate, context and max-output value is the gateway's; toolCalling is set only where blockrun recorded a structured tool_calls probe (GPT-6 family, Mistral Large 4). Media: xai/grok-imagine-video-1.5 in the video picker and VIDEO_PRICING ($0.08/s at the 480p default), plus a grok-video-1.5 pin. minimax/music-2.5 left the gateway catalog (remapped server-side to music-2.5+) and is off the music picker. Retired ids get the gateway's own redirect target as deprecated/fallbackModel, the tencent/hy3 convention: openai/gpt-5.3 -> gpt-5.2, grok-3/grok-3-mini/ grok-4-fast-* -> grok-4-1-fast-*, grok-2-vision -> grok-4.3. The grok-fast shorthand follows the redirect instead of chaining through it. Stale README rows for redirected ids removed. gpt-oss-120b: NVIDIA published its EOL on 2026-09-03 and the gateway marks it available:false. No default lands on it (free, FREE_MODELS and router-core's eco SIMPLE are nemotron-3.5-lightning); the comment claiming it had recovered is corrected and a test pins it off every default and generic alias. Bare shorthands are unchanged: laguna, mistral-large, tencent, hunyuan and qwen keep naming the free or cheaper model they always did. MODEL_ALIASES 276 -> 292; brand-numbers.json bumped in lockstep.
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info
📝 Walkthrough
Merge Risk: ⚪ Minimal · up to
|
| Check name | Status | Explanation |
|---|---|---|
| Description Check |  |
Check skipped - CodeRabbit’s high-level summary is enabled. |
| Title check | |
The title clearly and concisely describes the main change: synchronizing the hand-written model catalog with the live gateway catalog dated 2026-10-20. |
| Docstring Coverage | |
Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 2 functions across 4 files. (4 skipped: 4 … |
| Linked Issues check | |
Check skipped because no linked issues were found for this pull request. |
| Out of Scope Changes check | |
Check skipped because no linked issues were found for this pull request. |
Full details: Docstring Coverage
-
Explanation
Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 2 functions across 4 files. (4 skipped: 4 unsupported.)
✨ Finishing Touches
-
📝 Generate docstrings
-
- Commit to this branch
- Create a new PR
🧪 Generate unit tests (beta)
-
- Commit to this branch
- Create a new PR
-
- Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.
Comment @coderabbitai help to get the list of available commands.
VickyXAI
pushed a commit
that referenced
this pull request
Oct 20, 2026
DeepSeek V4.1 Flash added; free v4-flash pins follow the free default (#487). Catalog synced to the live gateway: six chat models, grok-imagine-video-1.5, retired ids follow the gateway redirect (#485). CI and publish on Node 24 (#488), scanner test reads .mjs chunks (#490), tsup 9 / vitest 5 (#486). dist/ rebuilt.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What changed
Added (registry
BLOCKRUN_MODELS, pickertop-models.json, explicit-pin aliases, README, SKILL.md)Pins:
gpt-6.1-sol,qwen3.8-max,hy4/hy4-preview,mistral-large-4,laguna-s,muse/muse-glimmer,grok-video-1.5. Barelaguna,mistral-large,tencent,hunyuan,qwenare deliberately unchanged (each names a free or cheaper model today; promoting them is a product call).toolCallingis set only where blockrun recorded a structuredtool_callsprobe (GPT-6 family, Mistral Large 4).free/muse-glimmer-30bis picker-listed but not in theFREE_MODELScascade (it returns empty content at smallmax_tokens).Media:
xai/grok-imagine-video-1.5added to the video picker andVIDEO_PRICING($0.08/s at 480p, 8s default).minimax/music-2.5dropped from the music picker (gone from the catalog; the gateway remaps it tomusic-2.5+). There is no hand-maintained speech list in this repo, sobytedance/seed-audio-1.0needs no change here.Retired ids follow the
tencent/hy3convention (entry kept,deprecated+ the gateway's own redirect target asfallbackModel, off the picker):openai/gpt-5.3→ gpt-5.2;xai/grok-3,grok-3-mini,grok-4-fast-*→grok-4-1-fast-*;xai/grok-2-vision→ grok-4.3. Thegrok-fastshorthand now namesgrok-4-1-fast-reasoningdirectly instead of chaining through the redirect. README rows for redirected ids removed.gpt-oss-120b: NVIDIA published its EOL 2026-09-03; the gateway marks it
available:falseand redirects tonemotron-3-super-120b. Checked: no default here lands on it (freealias,FREE_MODEL,FREE_MODELS[0]and router-core's eco SIMPLE primary are allnemotron-3.5-lightning). The alias comment claiming it had "recovered" is corrected, and a new test pins it off every default and generic alias.MODEL_ALIASES276 → 292;brand-numbers.jsonbumped in lockstep (blockrun'ssrc/app/brand/numbers.json/route.tsneeds the same 292).No
package.jsonchange, no version bump.dist/not rebuilt (left for the release commit).Sources
curl https://blockrun.ai/api/v1/models(2026-10-20).src/lib/models.tsandMODEL_REDIRECTSinsrc/app/api/v1/chat/completions/route.ts.Checks
npm run typecheck: passnpm run lint: passnpx prettier --check .: passnpm test: 1216 passed, 1 skipped (new2026-10-20 live catalog syncsuite)npm run build+ dist smoke: passSummary by CodeRabbit
New Features
Updates
Documentation