Google has launched Gemini 3.8 Flash, a new AI model arriving weeks after its predecessor. While keeping introductory pricing at $0.75 per million input tokens and $3.75 per million output tokens, Google warned that the model may use more tokens to maximize performance on complex tasks due to iterative tool calling and added reasoning steps.
The model outperformed previous systems on software engineering benchmarks like DeepSWE v1.1, as well as finance and legal agent tests. In response to safety concerns, Gemini 3.8 Flash includes built-in safeguards against CBRN and cyber offense misuses.
Alongside the general model, Google introduced Gemini 3.8 Flash Cyber under its Fairwind Program for select government and enterprise partners. The program provides specialized security tools, including the CodeMender agent, to autonomously detect and fix software vulnerabilities in critical infrastructure.
Why it matters
Highlights a shift toward iterative, agentic reasoning models that trade higher overall token usage for superior benchmark performance.
Offers developers high-speed, cost-effective coding capabilities while creating gated programs specifically for government and cybersecurity use cases.
Source: theverge.com
