What Happened
Google has officially released Gemini 3.8 and Gemini 3.8 Flash Cyber, marking its third Flash-class release in just six weeks. This ultra-fast rollout comes a mere three weeks after the debut of Gemini 3.7 Flash.
The new models focus heavily on "agentic workflows"—the ability of AI to complete complex, multi-step tasks autonomously—while matching the exact speed and low cost of their predecessor.
Why It Matters
The major breakthrough with Gemini 3.8 is its willingness to "work harder" on complex tasks. Rather than rushing to a quick response, the model is designed to execute extra reasoning steps and call external tools recursively to solve difficult puzzles.
This added diligence delivers massive performance gains across several key areas:
- Autonomous Engineering: On the DeepSWE v1.1 software engineering benchmark, Gemini 3.8 Flash solved complex coding problems end-to-end, outperforming much larger and more expensive competitor models.
- Specialized Expertise: The model showed massive leaps on the Vals Finance Agent and Harvey's Legal Agent benchmarks, highlighting its utility in highly technical industries.
- Creative Prototyping: Utilizing Google Antigravity, developers have already used Gemini 3.8 to generate fully playable 3D video games, a retro DOS version of Google Maps, and interactive topographic visualizers using real-world USGS datasets—all from single prompts.
Inside the Cyber Edition
Alongside the standard model, Google introduced Gemini 3.8 Flash Cyber. Released to trusted defenders via the Fairwind Program, this specialized model is built specifically to defend networks rather than exploit them.
The Cyber model boasts an impressive 70% success rate in discovering vulnerabilities across codebases spanning 20 different programming languages. On external patching benchmarks like CWE-Bench, Gemini 3.8 Flash Cyber performed on par with the industry's absolute best frontier models, but at a fraction of the running cost.
Furthermore, Google has built-in advanced safeguards against chemical, biological, and cyber-offensive misuses, alongside significantly boosting defenses against malicious prompt-injection attacks.
What's Next and Pricing
Google is keeping operational costs low to encourage immediate developer adoption. However, users should keep an eye on the calendar: the current introductory pricing is set to expire on December 31, 2026.
Starting January 1, 2027, standard pricing will adjust to $1.50 per 1 million input tokens and $7.50 per 1 million output tokens. For developers running highly efficiency-constrained workloads, Google notes that Gemini 3.7 Flash remains fully supported.
