Maestro Briefby Maestro Mojo

Gemini 4 Carbon: Watch the Access Gate and the Regular Price

Maestro Brief · Published by Maestro Mojo

2026-10-29

Maestro’s take. Put Gemini 4 Carbon on the evaluation list. Keep it off a delivery promise until you have access. For longer-term budgets, use the regular price rather than the introductory offer.

TL;DR

Google announced Gemini 4 Carbon on October 28, emphasizing coding, cybersecurity and sustained agent work.

The rollout starts with selected cyber defenders. Google plans to expand access through trusted partners before bringing Carbon to developers, businesses and consumers more broadly. The announcement gives no date for that wider availability.

Its announced API prices per million tokens are:

Token type Introductory → regular price
Input $2 → $4
Output $10 → $20

Google also announces a 95% discount on cached input tokens relative to the input-token price.

Why Maestro users care

Both access and price affect whether a model belongs in a production plan. A launch announcement cannot answer whether your team can run it next week. A temporary rate cannot establish the cost of a long-lived workflow.

Google reports stronger coding and cyber results. Its cited external cybersecurity benchmark runs were evaluated without guardrails. Those numbers do not establish the behavior, permissions or success rate of the deployed service your team might receive.

Maestro would keep the first comparison narrow: one representative coding task, a clear acceptance test, and the same tools and permissions as the current model. Count retries and review work. A lower token price is useful only when the finished job is worth buying.

One thing to watch

Watch for your actual access route, its terms and the offer’s end date. Once access arrives, compare cost per accepted result at both the introductory and regular rates before changing your default.

Published October 29, 2026 · Brief and Maestro commentary · Tags: models, costs, security

Original source: Google’s Carbon announcement, rollout details and pricing footnote.

MarkdownOpen in ClaudeOpen in ChatGPT