[VickyXAI]

What changed

Syncs the hand-maintained model lists with the live gateway catalog (2026-10-20). Two commits:

1. feat(models): sync with the live catalog

• src/pricing.ts: adds the 10 live ids that had no price — openai/gpt-6.1-sol ($2/$10), google/gemini-3.8-flash ($0.75/$3.75), deepseek/deepseek-v4-flash-vision-exp ($0.30/$1.20), qwen/qwen3.8-max ($2/$6), qwen/qwen3.8-flash ($0.15/$0.47), tencent/hy4-preview ($0.834/$2.501), xiaomi/mimo-v2.5 ($0.14/$0.28), poolside/laguna-s-2.1 ($0.09/$0.18), mistral/mistral-large-4 ($0.68/$2.09), nvidia/muse-glimmer-30b (free). Retired rows are kept on purpose. The table's own comments say it is the price of record for old session-cost records and hidden-but-served ids, so removing them would make old sessions show $0.
• src/ui/model-picker.ts: adds shortcuts for each new id, using the existing naming (gpt-6.1-sol, gemini-3.8-flash, qwen3.8-max, qwen3.8-flash, hy4, mimo-v2.5, deepseek-vision, mistral, mistral-large-4, laguna-s, muse-glimmer). Stale shortcuts are retargeted: hy3/tencent now point to tencent/hy4-preview, because hy3 was retired on 2026-09-24 and the gateway now redirects it to qwen3.7-flash. grok-4 now points to xai/grok-4.5, the same model as grok. grok-3, grok-fast and grok-4.1 now point to xai/grok-4.3, the model the gateway actually serves for those retired slugs. All other bare aliases are unchanged, and every free alias still resolves to a free model. Adds Mistral, Poolside and Cohere provider groups to the picker.
• src/free-models.ts / free-models.constants.ts: nvidia/nemotron-3-nano-30b was delisted on 2026-09-08, so it is removed from FREE_PREFERENCE_ORDER and FREE_MODEL_CONTEXT_WINDOWS and moved to LEGACY_FREE_MODEL_IDS. muse-glimmer-30b is added to the live free set, but not to the routing order, because it has not been measured on /v1/messages. The other dead free ids (mistral-nemotron, nemotron-nano-*, nemotron-3-super-*, step-3.7-flash) are only in LEGACY_FREE_MODEL_IDS, where they keep old sessions priced at $0 and keep stale pins behind the paid-approval gate. Nothing routes to them, so they stay. nano-30b now points to the free default.
• Tests: updated the pinned shortcuts and the free-chain fixtures, and added a regression test that checks every new id's price and shortcut.

2. fix(pricing): price retired Grok slugs at what the gateway charges
The gateway pins every retired xAI slug to grok-4.3 and bills $1.25/$2.50 (blockrun models.ts + model-maps.ts "Retired -> grok-4.3"). This table still had $3/$15 for grok-3 and grok-4-0709 and $0.2/$0.5 for the fast family. nvidia/kimi-k2.5 moves to the gateway's $0.60/$3.00. This is a separate commit so it can be dropped on its own if you'd rather keep the historical rates.

Sources

• GET https://blockrun.ai/api/v1/models and https://sol.blockrun.ai/api/v1/models, read 2026-10-20. Every new id is on both chains, and the only difference between the chains is azure/sora-2. No prices were inferred.
• Gateway catalog, used read-only: blockrun-deploy/src/lib/models.ts, model-maps.ts, and the chat route's MODEL_REDIRECTS, for the retirement and redirect facts.

Checks

• npm test (build + suite): 919/919 pass. One run had a timing flake in the unrelated Detach tool test, and it passed on rerun.
• npx tsc --noEmit: clean. node scripts/sync-brand-numbers.mjs --check: up to date. node dist/index.js --help: OK.
• The repo has no lint script.

No version bump and no CHANGELOG release header.