Google's Latest Gemini Flash Model Boosts Reasoning but May Inflate Bills
Google has introduced Gemini 3.8 Flash, which performs more reasoning steps and iterative tool calls than its predecessor. While per-token pricing remains unchanged, the model may consume more tokens at higher effort levels, potentially raising costs. Early analyses suggest a roughly 40% increase in cost per task compared to Gemini 3.7 Flash.
The new model's benchmark results span software engineering, finance, and legal domains, with Google highlighting particular strength in autonomous agent tasks. The company paired the release with safety measures targeting chemical, biological, radiological, and nuclear threats, as well as cyber offense.
The Fairwind Program, with 650 members including CrowdStrike and the Center for Internet Security, grants vetted government and partner access to the Cyber variant and CodeMender vulnerability-fixing agent. Consumer availability comes through Google AI Pro or Ultra subscriptions, while developers and enterprises can access the model directly.
The token consumption increase could meaningfully affect developers and businesses that rely on AI at scale, since a 40% cost rise per task may strain budgets even with