Google releases Gemini 3.8 Flash with iterative reasoning and tool use

Google has launched Gemini 3.8 Flash, arriving just a few weeks after its predecessor. The company reports that the new model performs more reasoning steps on complex tasks and calls tools iteratively.

Pricing and token consumption

The model maintains introductory pricing of $0.75 per million input tokens and $3.75 per million output tokens. However, Google notes that total expenses could increase because the model may use more tokens to maximize performance, especially at higher effort levels. Developers who want to minimize token usage can continue using Gemini 3.7 Flash.

Independent benchmarks show that Gemini 3.8 Flash achieves strong intelligence scores relative to its cost. Artificial Analysis noted that output tokens per task increased by about 30%, accompanied by more turns on agentic evaluations.

Benchmarks and agent capabilities

Google states that the model delivers significant improvements for software engineering and autonomous AI agents — software systems that can independently plan and execute multi-step tasks. Gemini 3.8 Flash outperforms its predecessor and other frontier models on the DeepSWE v1.1 software engineering benchmark, the Vals Finance Agent V2 benchmark, and Harvey’s Legal Agent benchmark.

The model also ships with safeguards against misuse in the domains of chemical, biological, radiological, and nuclear threats, as well as cyber offense.

Specialized versions and availability

Alongside the standard release, Google introduced Gemini 3.8 Flash Cyber through its new Fairwind Program. Limited to governments and trusted partners with roughly 650 members, including CrowdStrike and the Center for Internet Security, the program provides access to the specialized cyber variant and CodeMender, an agent designed to autonomously find and fix vulnerabilities.