3.40.0 — priced but unreachable: the catalog gets a front door

3.39.0 taught Franklin what the new models cost. It never taught anyone how to ask for them. That release added eleven chat models to src/pricing.ts and stopped there — no shortcut, no picker row, no router wiring. They were live, correctly priced, and reachable only by typing the full provider/model id from memory.

38 new shortcuts (107 → 145)

family	now reachable as
GPT-5.6 pro reasoning tier	sol-pro, terra-pro, luna-pro
GPT-5.5 Pro / ChatGPT default	gpt-5.5-pro, chatgpt / instant
Gemini 3.6 Flash + Flash Lite line	flash (now 3.6), flash-lite, gemini-3.1-flash-lite
Qwen3.7 Plus / Flash	qwen-plus, qwen-flash
Tencent, Xiaomi (new providers)	hy3 / tencent, mimo / xiaomi
GPT-4o + 4.1 mini/nano, o3-mini	4o, gpt-4o-mini, gpt-4.1-mini, o3-mini

Terra Pro ($1/$6) and Luna Pro ($0.1/$0.6) undercut their own base tiers while adding pro reasoning mode. gpt stays pinned to Sol — bare aliases follow the gateway's flagship, not the cheapest sibling.

New arrivals: zai/glm-5.3 and xai/grok-imagine-video-1.5 ($0.08/s) in VideoGen.

The router's LLM classifier had been dead since 2026-07-27

Its default was nvidia/qwen3-next-80b-a3b-instruct — NVIDIA's EOL'd id, which the gateway rides on nemotron-3-super-120b. That substitute opens with "Okay, let's see. The user wants to…" and never reaches a verdict inside the classifier's token budget, so every classification failed the strict parse and fell through to keyword-only routing. Invisible by design, which is why it went three weeks unnoticed. Re-probed the free pool on the real classifier prompt — only nemotron-3-nano-omni returns one bare word. It's the new classifier.

That model also started serving itself again (it was pooled behind gpt-oss-120b in August), so it gets the omni alias, a free picker row, and second place in every free chain. nvidia/step-3.7-flash stays deliberately unaliased — $0 and in the catalog, but still served by a substitute.

The hand-curated tables the new aliases exposed

• xai/grok-4.5 was missing from the vision allowlist. It's what grok resolves to and it is vision-capable, so every image turn on the xAI flagship was rerouted to a sibling the user never chose. Eleven more were missing. A new test pins the invariant: every bare flagship alias must be vision-listed.
• Context windows for grok-4.5 (500K), grok-4.3 (1M) and glm-5.3 (1M) were absent and match no inference pattern — a cold cache used the blind 128k default and compacted up to 8× too early. 12 entries added.
• Model-family guidance matched bare glm from the GLM-4.x era, telling a 1M-context flagship to make ONE tool call and stay under 300 words. GLM-5.x moves to balanced guidance.

Chain parity

GLM-5.3 is Base-only — Solana lists 92 of Base's 93 models and 5.3 is the one it's missing (prices otherwise identical). So glm stays pinned to 5.2: a bare alias has to resolve on both chains, or every Solana user gets an HTTP 400. 5.3 keeps its explicit alias and picker row.

Picker holds its 24-row cap — new models displaced old ones rather than growing the list. 647 local tests pass; every model id in this release was probed against the live gateway (402 = exists, 400 = unknown), and free models were probed for self-service rather than assumed.