Vibeleaderboard
← All Intel
Intel / article

Gemini 4 Carbon now available on AI Gateway

Source
vercel.com
Author
Jerilyn Zheng
Date

Context

Vercel's AI Gateway now serves Gemini 4 Carbon under the model name google/gemini-4-carbon, at 50% off through November 30, 2026 (changelog). AI Gateway is a single API that forwards requests to many providers' models and keeps usage and cost in one place. Vercel's post gives no per- prices, but the end date matches Google's introductory period: its release post sets Carbon at $2 per million input tokens and $10 per million output tokens, and a footnote on Google's page says $4 and $20 apply after November 30. Vercel says the gateway passes provider pricing through with no markup and no platform fee on , so the discount appears to be Google's launch price rather than a separate Vercel offer, though the post does not say so directly.

What the gateway adds is the plumbing around the model. Code already written with Vercel's AI SDK switches to Carbon by changing the model string. The gateway records usage and cost per request, can retry or fail over to keep requests running when a provider has trouble, and supports budgets on API keys, routing rules and Zero Data Retention. For coding agents, running npx vercel ai-gateway setup with the latest Vercel CLI connects tools such as Claude Code, Codex and Cursor to the gateway, after which Carbon can be picked inside the . Vercel describes Carbon as built for complex software engineering, long agentic work and cybersecurity defense, with a of 1 million tokens.

Terms in this piece · Glossary
  • token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
  • inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
  • AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
  • context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
More from Jerilyn Zheng
Recommended reads
Comments

Checking sign-in…

Loading comments…