chore(ai): regenerate the model catalog from live provider catalogs

[eliebak]

Part of ENG-5435.

Context

The committed models.generated.ts snapshot has drifted since #1445. Release builds regenerate this file from live catalogs anyway; committing the refresh keeps the checked-in snapshot from drifting further and keeps local keyless test runs representative. Follow-up PR adds a daily workflow to automate this.

Changes

Snapshot generated for commit 4332c7a (1252 -> 1260 models):

• Added z-ai/glm-5.3 (OpenRouter, Prime Inference), deepseek/deepseek-v4-flash-vision-exp (OpenRouter, Vercel AI Gateway, OpenCode Go), thinkingmachines/inkling:free + inkling-small:free, and stealth previews stealth/ox-alpha / ox-alpha-free / x-preview-f-free.
• Vercel AI Gateway renamed its grok vendor prefix from xai/ to spacexai/ (13 ids). Anyone pinning a vercel-ai-gateway grok id will need the new prefix.
• Repricing: gpt-5.6-sol 5/30 -> 2.5/15 (2/10 on OpenRouter), gemini-3.6-flash and gemini-flash-latest halved to 0.75/3.75, gemini-flash-lite-latest raised to 0.3/2.5, gpt-5.6-terra now free-preview (0/0), smaller adjustments on kimi-k2.7-code, minimax-m2.5, mistral-small-3.2, muse-glimmer-30b.
• Retired deepseek-v4-flash-free (OpenCode).

Validation

• Ran npm run generate-models in packages/ai; all four sources (models.dev, OpenRouter, Vercel AI Gateway, Prime Inference) fetched successfully. Prime loaded 105/105 models from the live catalog.
• Verified no provider section was emptied by a failed fetch (per-provider counts match live catalogs; total went 1252 -> 1260).

🤖 Generated with Claude Code

---

Note

Medium Risk
Generated catalog only, but Vercel AI Gateway grok IDs change from xai/ to spacexai/, which will break anyone pinning those IDs. Cost and max-token updates also affect billing estimates.

Overview
Refreshes the committed models.generated.ts snapshot from live provider catalogs (1252 → 1260 models).

Adds GLM 5.3 (OpenRouter, Prime Inference), DeepSeek V4 Flash Vision Exp, Inkling free routes, and Ox Alpha / preview free models. Removes OpenCode deepseek-v4-flash-free.

Breaking for pinned IDs: Vercel AI Gateway grok models rename from xai/ to spacexai/ (13 IDs).

Also picks up repricing (notably gpt-5.6-sol and Gemini Flash variants) plus various context/max-token and cache-read tweaks.

^{Reviewed by Cursor Bugbot for commit 4332c7a. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Regenerate models.generated.ts from live provider catalogs

• Refreshes the generated model catalog across multiple providers with new models, repricing, and updated token/context limits
• Adds new models: GLM 5.3, DeepSeek V4 Flash Vision Exp (text+image), thinkingmachines/inkling, stealth/ox-alpha, and several free-tier routes
• Renames all xai/* routes to spacexai/* on the Vercel AI Gateway provider
• Reprices numerous models including gpt-5.6-sol, gemini-3.6-flash, deepseek-v4-flash, and others across GitHub Copilot, Google, OpenCode, OpenRouter, and Vercel AI Gateway
• Risk: any downstream code or configs referencing old xai/* route names on Vercel AI Gateway will break; check references to xai/ routes and update to spacexai/

^{Macroscope summarized 4332c7a.}

[sethkarten]

Design and implementation reviewed. This is a clean generated snapshot refresh with the required changelog fragment. The provider-ID changes are intentional upstream catalog changes; aliases or copy-only tests would add brittle compatibility scaffolding. Local AI validation passed (338 tests, 50 files), root checks passed, and the Linear link is now green.