Google launches Gemini 3.8 Flash, more powerful but costlier

Gemini 3.8 Flash arrives just weeks after its predecessor, signaling a fast release cadence focused on pushing performance. Google frames the upgrade around 'working harder,' meaning the model performs more reasoning steps on complex tasks and calls tools iteratively. This approach delivers gains on benchmarks, but it comes with a cost caveat: the model may use more tokens per task, potentially raising expenses even though per-token prices remain flat.

Early assessments reinforce this trade-off. Artificial Analysis estimates the model is ~40% more expensive per task than Gemini 3.7 Flash, driven by a 30% increase in output tokens per task, though it still calls it the 'cheapest we've measured at this level of intelligence.' Aigora.ai CEO John Ennis compared its coding quality to Anthropic's Opus 5, but at 'a fraction of the cost and super fast.' Google's claims of outperforming rivals on the DeepSWE v1.1, Vals Finance Agent V2, and Harvey's Legal Agent benchmarks position the model as a strong choice for software engineering and autonomous agents.

Google also paired the launch with a security-focused offering under its new Fairwind Program. Restricted to governments and trusted partners, it provides access to Gemini 3.8 Flash Cyber and the CodeMender agent, which can autonomously find and fix vulnerabilities. The program's membership includes CrowdStrike and the Center for Internet Security, and the model ships with safeguards against misuse in CBRN and cyber offense domains, reflecting an emphasis on safety in high-stakes deployments.