Gemini 2.5 Pro and Flash auto-reroute to Flash-tier models Dec 1
The swap changes your model, price and parameters at once — plus Vercel's security beta, LLM 0.37's new OpenAI models, and CodeQL for Python 3.15.

Copy markdown
Your gemini-2.5-pro string now points at a Flash model
From Dec 1, requests to `gemini-2.5-pro` route to `gemini-3.8-flash` and `gemini-2.5-flash` routes to `gemini-3.5-flash-lite` — billed at the new models' rates. If either string is hardcoded in production, your quality and cost profile shifts with no code change on your end; pin an explicit model version before the switch.
The reroute ships breaking parameter changes too
It isn't just a rename: `thinking_budget` is replaced by `thinking_level`, several sampling parameters and prefilled model turns are removed, and function calling now enforces `thought_signature` validation. Run your Gemini calls against the 3.x models now or they may start erroring on Dec 1.
Vercel adds push-time Security Review with agent autofix
Vercel's Security Review (public beta, Nov 3) scans every push to your GitHub repos for vulnerabilities and lets Vercel Agent open the fix PR. The same release adds standalone Vercel CI that no longer needs a linked project, plus a `vercel doctor` command that checks your CLI, login, team, and `vercel.json`.
LLM 0.37 adds the newest OpenAI models, halves its default cost
Simon Willison's `llm` CLI 0.37 adds `gpt-6.1-sol` and `gpt-6.1-terra`, and makes `gpt-6-luna` the default — which he says costs half what the prior default (`gpt-5.6-luna`) did. A new `llm openai endpoint add` command wires up any OpenAI-compatible endpoint, and embeddings now accept multimodal content types.
Vercel MCP: connect with no pre-registration, deploy from ChatGPT
Vercel MCP now accepts Client ID Metadata Documents as OAuth client IDs, so MCP clients can connect without being pre-registered. Deployments also render as interactive cards inside MCP Apps clients like ChatGPT, where you can track status, read build logs, and debug without leaving the chat.
CodeQL and Dependabot catch up to Python 3.15
GitHub shipped CodeQL 2.27.3 with Python 3.15 feature support and JSON data extensions, and Dependabot now supports Python 3.15 (both Nov 3). If you've already moved to 3.15, your code scanning and automated dependency bumps work again.