Google announced Gemini 4 Carbon on October 28, 2026, but the announcement does not describe immediate general access. It says availability starts this week with a subset of cyber defenders, followed by additional trusted partners through Fairwind. Wider availability for developers, businesses and consumers remains forthcoming.
Google’s Carbon announcement, rollout description and pricing footnote
Confirm access before committing a project
For a team planning an integration, the first deliverable should be an access record: the exact model, the approved account, the product surface and a successful authorized test. Keep a proposed future model separate from the model currently serving your application. A release headline alone cannot complete that record.
DeepMind’s Fairwind page describes managed access for vetted partners, prioritizing governments and critical infrastructure. It offers an application route and says partners cannot share, redistribute or sell program access. Applying is a request for review, not evidence of admission. The page’s detailed model sections still name Argon; use the new Carbon announcement for Carbon’s rollout statement.
Google DeepMind Fairwind Program: application, governance and managed access
Put both price periods in your budget
Google announces introductory Carbon rates of $2 per million input tokens and $10 per million output tokens. Its footnote gives subsequent rates of $4 and $20 respectively. The announcement does not state an expiry date for Carbon’s introductory period. Do not import a date from another model’s offer.
Consider a fictional workload with two million uncached input tokens and 100,000 output tokens. Applying the introductory rates gives $4 for input plus $1 for output, or $5. Applying the subsequent rates to the same quantities gives $8 plus $2, or $10. This is a token-price exercise, not a measured Carbon bill or a promise of access.
Use two columns in a planning sheet: introductory and subsequent pricing. Add your own expected quantities, then a separate allowance for repeated attempts and longer responses. Keep optional services and other operating costs outside this narrow token subtotal until their applicable terms are verified.
Define what would justify a switch
While waiting for confirmed access, choose a small task you already understand. Write down its inputs, expected output, acceptance criteria and current cost. A useful trial might ask for a change in a disposable copy of a project, followed by tests and human review. Keep that trial bounded so a longer run does not become an unexamined production commitment.
Record the actual model used, elapsed time, billable usage, failed attempts and review effort. Compare the accepted result with your existing workflow. A cheaper token rate can still produce a more expensive accepted task if it requires more attempts; an impressive demonstration can still fail your specific acceptance criteria.
The practical next step is therefore to prepare the evaluation and both budget scenarios, then update them when your account’s access and commercial terms are confirmed. Keep the rollout decision and the purchasing decision as two separate approvals within your own project.
Check the planning distinction
