Google’s Gemini 4 Carbon is coming first to cyber defenders
Google’s newest frontier model promises stronger coding and cybersecurity help. Access starts with selected defenders, not everyone with a Gemini subscription.
- Maturity
- Experimental, stage 1 of 4
- Support
- 3 sources · institution
- Evidence detail
- How we know ↓

Google announced Gemini 4 Carbon on October 28, with selected cyber defenders due to receive access this week. Trusted partners follow through Fairwind; broader access for developers, businesses and consumers remains ahead. Google
Google reports stronger coding, sustained engineering and reliability. These are company findings, not independently established outcomes.
Capability is not a finished service
Its model table lists Carbon at 71% on CWE-bench versus Argon’s 68%, and 88.2% on SEC-bench Pro versus 83.4%. These are test scores, not the share of all software flaws a model can find or fix. Carbon also trails some listed competitors on FrontierSWE 2 and OSWorld-2.1 offline. Model comparison
The cybersecurity benchmarks were run without guardrails. Deployed restrictions may therefore change what a model will do.
Fairwind vets partners and restricts use to defensive or research purposes. Its program page still describes Argon in places; use the Carbon announcement for the new rollout, rather than assuming every partner already has access. Fairwind
For everyday people, the hopeful possibility is software that gets repaired more effectively. That benefit still needs evidence: reproducible findings, reviewed repairs and a clear record of what remains unresolved. A stronger benchmark result is not a reason to buy a subscription expecting immediate access.
How we know3 sources · checked 2026-10-28 · no corrections
Original sources
- Google · Carbon announcement ↗ · institution
- DeepMind · Gemini ↗ · institution
- DeepMind · Fairwind ↗ · institution
Institutions: Google · Google DeepMind
- Maturity
- Experimental
- Event date
- 2026-10-28
- Source published
- 2026-10-28
- Captured
- 2026-10-28
- Last source review
- 2026-10-28
- Editorial method
- AI-assisted source review
Bright compared this account with the linked original and supporting sources and kept reported, budgeted, projected, and observed claims distinct. Bright did not independently audit the underlying records.
Maturity describes the tested or operational setting. Confidence describes support for the particular claim; one does not determine the other.
Revision & correction history
2026-10-28T21:10:00.000Z · Original reporting on Carbon’s selective rollout.
No corrections recorded.
Keep exploring
Explore the shared question in another setting. These connections do not imply replication.
