Google Tests Gemini 3.8 Flash Internally

A preliminary version of Gemini 3.8 Flash is already circulating among Google employees. With closely spaced releases and a bet on faster, cheaper models, the company is adjusting its offerings in response to OpenAI and Anthropic, without any guarantee on the date of a public launch.

An Accelerating Timeline, Without a Guaranteed Date

Google rolled out Gemini 3.6 Flash in July, followed three weeks later by 3.7 Flash. Sundar Pichai indicated during the second-quarter earnings call that the company aims for an almost monthly launch cadence for its models. Last month, a new AI model was launched. The early access to Gemini 3.8 Flash by employees suggests that a public launch could be considered soon, though there is no certainty. Industry observers claim to have spotted 3.8 Flash on Arena AI, a benchmarking platform that Google has previously used to test upcoming versions.

What Teams Are Already Testing Internally

Google employees are currently testing Gemini 3.8 Flash, in a preliminary form revealed internally. This model, called "Gemini 3.8 Flash Preview," is accessible to staff via Jetski, the internal coding platform. An employee involved in these trials believes that the model already appears significantly better than 3.7 Flash, while noting that it is too early to make a definitive judgment. Google has declined to comment on these tests.

Focus on Flash Models and the Race for Agents

Gemini 3.5 Pro is still not available, and Google no longer has a leading model in the face of Anthropic and OpenAI. The company is now betting on its Flash models, touted as faster and cheaper for users. Google describes these models as "work models" suited for coding and executing agents, uses that often consume tokens more quickly than chatbots, as clients closely monitor their AI spending. Agentic assistants are becoming a major trend: Google launched the Gemini Spark agent earlier this year, while Meta is internally testing an agent called "Hatch."