Gemini 3.8 Flash: Stable Token Cost, Potentially Higher Usage Fees

Google is launching Gemini 3.8 Flash at the same token price as 3.7, while warning that the model may consume more tokens. The company highlights superior results across several benchmarks and introduces a cyber variant reserved for a limited access program.

The cyber variant goes through a restricted program

Google has simultaneously launched Gemini 3.8 Flash Cyber and the Fairwind Program. This initiative is limited to governments and trusted partners and includes 650 members, among them CrowdStrike and the Center for Internet Security. It provides access to 3.8 Flash Cyber and the CodeMender agent, which Google claims can autonomously find and fix vulnerabilities to protect critical infrastructures, public services, and national security. Google also states that Gemini 3.8 Flash incorporates protective measures against abuses in chemical, biological, radiological, and nuclear (CBRN) domains as well as against cyberattacks.

Unchanged pricing, variable billing based on token usage

The introductory price remains at $0.75 per million input tokens and $3.75 per million output tokens, the same as 3.7 Flash. However, Google warns that the model may use more tokens to maximize performance, particularly at high effort levels, which could increase the bill. Developers can continue to use 3.7 Flash if they want to minimize this consumption. Artificial Analysis estimates that 3.8 Flash is the cheapest measured at this level of intelligence, while reporting about a 40% increase in cost compared to 3.7 Flash, attributed to 30% more output tokens per task and more rounds during agent evaluations.

Benchmarks and comparisons against Anthropic

Google emphasizes improvements for software engineering and autonomous agents, with superior results on DeepSWE v1.1 compared to its previous model and other benchmarks, including Anthropic's Fable 5. According to tests, the latest model also scored better than its rivals on Vals Finance Agent V2 and Harvey’s Legal Agent. Anthropic updated Fable 5 earlier this week and announced enhanced performance at a lower cost due to a decrease in the price of cached data usage. John Ennis, who leads Aigora.ai, believes that 3.8 Flash delivers Opus 5 coding quality at a fraction of the cost and very quickly, citing potential uses such as creating remotion videos.

Availability and functional positioning

Gemini 3.8 Flash was launched a few weeks after 3.7 Flash. It is accessible to consumers through Google AI Pro or Ultra subscriptions, as well as to developers and businesses. Google claims that the model "works harder" than its predecessor by multiplying reasoning steps on complex tasks and iteratively calling tools.