Google is preparing to release a new AI model built specifically for software development, giving developers a faster, cheaper option tuned for writing and editing code. The release arrives as Google DeepMind reshapes its leadership and sharpens its focus on the coding race that has come to define the AI market in 2026.<\/p>\n
The model in question is Gemini 3.8 Flash, internally known as “Skimaki.” According to people familiar with the matter, it could be released as early as Wednesday, just one day after the details first surfaced. Internal testers describe its coding capabilities as a significant upgrade over earlier versions.<\/p>\n
The AI industry has spent most of 2026 fixated on coding. Agentic coding, where AI tools take multi-step actions to build and debug software, became the dominant trend of the year, and token-heavy agentic workflows made speed and cost as important as raw model size. Google finished 2025 near the top of the AI rankings, then found itself pressured by platforms that emphasized fast code generation. A coding-first release is a direct response to that pressure.<\/p>\n
Gemini Flash models are not designed to be the largest or most powerful systems on the market. They are built to run faster and cost less to operate, two qualities that have grown more valuable as developers route requests through multiple models at once.<\/p>\n
The coding model was evaluated on an internal Google tool called Jetski. According to people familiar with the testing, coders who tried Skimaki said they preferred it to Anthropic’s Claude Opus for everyday development work. That preference, if it holds up in broader use, would give Google a stronger foothold in the developer tools market, where Anthropic has been a leading name.<\/p>\n
The timing of the release is closely tied to changes at Google DeepMind. Demis Hassabis, who founded DeepMind and served as its CEO, stepped aside from the chief executive role early in August 2026. Insiders had reported for roughly a year that he was preparing to scale back his operational duties. In his new role, Hassabis is focused primarily on scientific research, the area where he has historically preferred to spend his time.<\/p>\n
Hassabis, a Nobel Prize winner in chemistry, had publicly expressed caution about rushing frontier AI systems to market. His exit from the CEO seat clears the way for a faster, more product-driven approach at DeepMind.<\/p>\n
Google co-founder Sergey Brin has taken a more active interest in AI and has pushed to double the pace of Gemini model releases. With Hassabis concentrating on research, the operational direction of Gemini development now sits closer to Brin and the broader Google leadership team, which has signaled that shipping competitive coding models is a near-term priority.<\/p>\n
Frontier AI models, the largest and most parameter-heavy systems, now tend to arrive with extensive safety reviews and public scrutiny before launch. Flash models take a different approach. They prioritize latency, cost, and developer accessibility over raw scale, making them well suited to the agentic coding workflows that have taken hold across the industry.<\/p>\n
Gemini 3.8 Flash, also called “Skimaki,” is a Google AI model built with a focus on coding. It belongs to the Flash line, which prioritizes speed and lower cost over maximum model size.<\/p>\n
The model could be released as early as Wednesday, according to people familiar with the matter. The release was first reported the day before, on a Tuesday.<\/p>\n
Coding has become the most active area of competition in AI for 2026, driven by the rise of agentic coding tools. A coding-first model lets Google compete directly with Anthropic’s Claude Opus and other developer-focused systems.<\/p>\n