fix(models): sync curated picker with the live BlockRun catalog - #32
Merged
VickyXAI merged 2 commits intoAug 31, 2026
Merged
Conversation
Mirror ClawRouter's current top-models.json curation plus the new upstream models it has not picked up yet: - add zai/glm-5.3 and glm-5.3-flash - add openai/gpt-5.5-pro - add google/gemini-3.6-flash and gemini-3.5-flash-lite - add qwen/qwen3.8-flash, tencent/hy3, xiaomi/mimo-v2.5 and -pro - add deepseek/deepseek-v4-flash-vision-exp - add the current free tier: nemotron-3.5-lightning, nemotron-3-nano-30b, laguna-xs-2.1, north-mini-code, nemotron-3-ultra-550b, llama-3.2-11b-vision - retire free entries no longer in the live catalog: seed-oss-36b, mistral-nemotron, step-3.7-flash, nemotron-nano-9b-v2, nemotron-nano-12b-v2-vl Legacy models still routed by the gateway (o1/o3, gpt-4o, gpt-4.1, gemini-2.5, ...) are deliberately left out of the picker to keep the keyboard curated; they remain routable by id.
… order Review follow-up on the catalog sync. Three things the sync missed, all of which the mirror guard would have caught had it been able to run: * blockrun/free/deepseek-v4-flash is dead. NVIDIA EOL'd it 2026-08-12 with 410 Gone (blockrun #367); the gateway carries it hidden:true and server-redirects calls to a healthy free model, and ClawRouter pulled it from the picker, the FREE_MODELS cascade and the router fallbacks — top-models.test.ts asserts it stays out. A [FREE] button for it answers from a different model and silently defeats /exclude, the same shape as the qwen3-next entry already in retired_models. Retired here too, and removed from the four surfaces that named it. * Picker order had drifted from top-models.json at two pairs: qwen3.8-flash sat ahead of mimo-v2.5 (39 vs 34) and deepseek-v4-pro ahead of deepseek-v4-flash-vision-exp (41 vs 40). * gemini-3.6-flash, gemini-3.5-flash-lite, tencent/hy3 and mimo-v2.5-pro are live in the BlockRun catalog but not yet in top-models.json, so they belong in POST_TOP_MODELS_ADDITIONS rather than failing the membership check. test_curated_picker_catalog_mirrors_clawrouter_top_models only runs where a sibling ClawRouter checkout exists, so CI skipped it and it was red locally on both this branch and main. With a sibling checkout: 83 passed, 0 failed. Also swept the doc surfaces nothing enforces: README's picker table still listed the five free models this branch retired and none of the eleven it added, and docs/03's in-fence model count was frozen at 71 (the brand-sync script skips fenced code blocks by design, so it never self-heals). SKILL.md's "free NVIDIA models" no longer holds — north-mini-code is Cohere's and laguna-xs-2.1 is Poolside's.
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
2 participants
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.
Summary
Sync the curated picker catalog with the live BlockRun catalog, mirroring ClawRouter's
top-models.jsoncuration plus the new upstream models it has not picked up yet.Added:
zai/glm-5.3,zai/glm-5.3-flashopenai/gpt-5.5-progoogle/gemini-3.6-flash,google/gemini-3.5-flash-liteqwen/qwen3.8-flashtencent/hy3xiaomi/mimo-v2.5,xiaomi/mimo-v2.5-prodeepseek/deepseek-v4-flash-vision-expnemotron-3.5-lightning,nemotron-3-nano-30b,laguna-xs-2.1,north-mini-code,nemotron-3-ultra-550b,llama-3.2-11b-visionRetired (no longer in the live catalog):
free/seed-oss-36b,free/mistral-nemotron,free/step-3.7-flash,free/nemotron-nano-9b-v2,free/nemotron-nano-12b-v2-vlDeliberately omitted: legacy models still routable by id through the gateway (o1/o3/o3-mini, gpt-4o/-mini, gpt-4.1 trio, gpt-5-mini, gpt-5.2/-pro, chat-latest, claude-opus-4.5, claude-sonnet-4.5, gemini-2.5 trio, qwen3.7-plus/flash, gpt-5.6-*-pro) — kept out of the picker to avoid flooding the Telegram keyboard.
Tests
test_provider_template.py(free tail order mirrorstop-models.json, retired-model guard),test_telegram_labels.pyfree sample, SKILL.md free list, docs count.python3 -m pytest: 82 passed, 4 skipped.Verification on testbed
hermes-clawrouter doctor: all checks pass[FREE]entrieszai/glm-5.3→ HTTP 200, replied "OK"