Google’s Gemini 3.8 Flash is built for agents, while its Cyber twin hunts vulnerabilities

Google’s Gemini 3.8 Flash is built for agents, while its Cyber twin hunts vulnerabilities

Taryn Plumb

5:02 pm, PT, September 2, 2026

Google continues to release Flash models, with two new versions of its 3.8 Flash announced on Wednesday.

Variants of 3.8 Flash

The variants include:

  • A standard Flash, a “workhorse” model for agentic tasks, software development, and multi-step reasoning.
  • Flash Cyber, optimized for vulnerability detection and mitigation.

Google CEO Sundar Pichai highlighted that 3.8 Flash delivers “significant leaps” from 3.7 Flash in software engineering, agentic tasks, and multi-step reasoning.

Flash Cyber’s Capabilities

Flash Cyber is Google’s most capable cybersecurity model, achieving 86.2% on the CyberGym cybersecurity benchmark and 47.2% on CWE-Bench, which evaluates AI patching abilities. In an internal benchmark, it discovered vulnerabilities across 20 programming languages with a success rate of over 70%.

Availability and Pricing

The 3.8 Flash is available now in Gemini Enterprise. Developers can try it out in the Gemini API via Google AI Studio, Google Antigravity, Android Studio, or generate UIs in Stitch. Pricing is $0.75 per million input tokens and $3.75 per million output tokens.

Key Features of 3.8 Flash

  • Works harder with greater diligence, executing extra reasoning steps but potentially using more tokens.
  • 1M-token input window and 64K-token output limit.
  • Can ingest text, images, audio, video, and PDF files.

Benchmarks and Specialized Knowledge

The model was evaluated across numerous benchmarks and performed well in:

  • Coding and multimodal capabilities.
  • Computer use and long-context knowledge work.
  • Scientific reasoning.
  • Specialized knowledge domains requiring in-depth analysis and reporting.

For example, Gemini 3.8 Flash:

  • Outperformed its predecessor and other frontier models on Vals Finance Agent V2 for finance and Harvey's Legal Agent Benchmark for law.
  • Achieved 54.9% on Humanity’s Last Exam (HLE)-Verified, demonstrating multi-step reasoning across diverse subjects.

Examples of 3.8 Flash Applications

  • Built a game with a simple prompt using looping techniques in Google’s Antigravity platform.
  • Created a fully-functional DOS version of Google Maps with interactive locations, directions, and street views.
  • Developed a 3D visualizer that decomposed devices into layers for inspection.
  • Generated a topographic map of famous geographical sites based on real datasets from the U.S. Geological Survey.