Google announced Gemini 4 Carbon on October 28, targeting sustained software-engineering workflows and cybersecurity defense. Google says its internal teams have seen Carbon investigate engineering problems for days, including tracing and patching difficult bugs across large codebases.
The practical details
- Coding gains: Google reports 65.2% on Terminal-Bench 4.0 (AA), versus Argon’s 57.1%. That is a benchmark result, not a production-reliability guarantee.
- Restricted access: rollout starts with selected cyber defenders, followed by additional trusted partners through Google’s Fairwind Program. Broader developer and consumer access has no firm date.
- Announced pricing: introductory rates are $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 afterward. The announcement does not specify when introductory pricing ends.
The internal engineering examples are Google’s reports, not an independent evaluation of how Carbon will behave on another team’s repository. Announced pricing also should not be confused with general availability.
Why it matters for builders
The interesting shift is persistence: agents that keep investigating stubborn problems instead of stopping after one plausible answer. The useful outcome is a change that survives verification, not simply a longer run.
For a small app studio such as SunMarc, a useful evaluation—once access opens—would be a bounded bug-fixing task, judged by passing tests and reviewable patches. Compare accepted fixes, review effort, elapsed time and total cost against the current workflow.
That is a proposed evaluation, not an announced SunMarc integration or a claim that we have tested Carbon. Our coverage of Kimi K3.1 Pro for longer coding tasks explores a related evaluation question with different access conditions.