Back to Timeline

Event Summary

Google introduced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber on September 2, 2026. The standard model is available through Google products and APIs at introductory pricing; the Cyber model is restricted to trusted defenders through the Fairwind Program. Google reported coding, vulnerability-discovery, patching, and Chrome security results, which are not independent replications of the vendor's full evaluation program.

Context & Narrative

Google says 3.8 Flash improves long-horizon software engineering and that its higher effort settings can consume more tokens. For Flash Cyber, Google reports a 47.2% pass@1 on CWE-Bench and internal Chrome outcomes, while saying the model is distributed only to trusted defenders because of more permissive cyber mitigations. EpochArc records the release and access model as facts and labels all performance and operational results as Google's reports.

Key Findings

  • Fact Grade B

    Google introduced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber on September 2, 2026; the latter is restricted through the Fairwind Program.

    Sources [1][2]
  • Impact Grade B

    Google reported strong software-engineering scores, a 47.2% CWE-Bench pass@1 for Flash Cyber, and internal Chrome patching results.

    Sources [1]
  • Limitation Grade B

    Google says complex-task performance can require higher token use, while Flash Cyber remains limited to trusted defenders rather than broad public access.

    Sources [1]

Impact Assessment

  • Capability Leap +1 · Short-term

    Adds a low-cost general model and a gated cyberdefense variant with Google-reported software and patching performance.

    Affected Groups: software developers, cybersecurity defenders, enterprise teams

  • Risk Creation -1 · Medium-term

    The Cyber variant's more permissive safeguards create dual-use concerns that Google addresses through trusted-defender distribution.

    Affected Groups: critical infrastructure operators, security community

Consensus & Sources

Significance L1
Category Capability Breakthrough / Safety & Ethics
Consensus Emerging Consensus
Impact Index 5/10