Google eyes coding comeback with new Gemini 3.8 Flash model

The move comes as AI coding tools become a key battleground, with Google seeking to strengthen developer adoption, accelerate model development and compete more effectively against increasingly capable offerings from rival AI firms.

Google is reportedly preparing a new Gemini model focused on improving coding performance, as competition in the AI coding market intensifies. Internally referred to as “Skimaki”, the model is expected to be introduced as Gemini 3.8 Flash and could arrive as early as September 2, according to reports.

The development comes as Google looks to regain momentum in AI-powered coding, an area where rivals such as OpenAI and Anthropic have built a strong position. Google has been testing the new model internally since August, with coding performance emerging as one of its key priorities.

Focus shifts towards coding performance

The new model is reportedly being evaluated through Google’s internal coding platform, Jetski, where some engineers have found its performance competitive with Anthropic’s coding-focused models. The reported results indicate that Google is seeking to make its Flash series more capable while retaining the speed and cost advantages associated with smaller AI models.

Flash models are designed to deliver faster responses and lower operating costs compared with larger Pro models. This makes them particularly relevant for coding assistants and AI agents, where developers may need to generate, test and modify code repeatedly.

Google’s recent model development has also reflected this strategy. The company reportedly released Gemini 3.6 Flash in July, followed by Gemini 3.7 Flash only a few weeks later. Meanwhile, work on a Gemini 3.5 Pro model was reportedly delayed after internal testing showed that it did not deliver sufficient improvements over existing Flash models.

Google’s wider AI organisation has also undergone changes as the company attempts to accelerate development. Demis Hassabis has stepped back from day-to-day responsibilities at Google DeepMind, with Koray Kavukcuoglu taking a larger operational role. The company has also recruited Barret Zoph, a former OpenAI post-training leader and Thinking Machines Lab co-founder, to work on research and reinforcement learning.

Pressure builds across AI coding market

The push comes at a time when AI coding tools are becoming an increasingly important battleground for technology companies. OpenAI and Anthropic have gained traction with developers through models designed for software development and autonomous coding tasks, increasing pressure on Google to improve Gemini’s capabilities.

Google’s Gemini 3.0, launched in November 2025, initially strengthened the company’s position in the AI race. However, subsequent advances from competitors have increased the need for faster model development and stronger specialised capabilities.

The company is also reportedly working on Gemini 4, which has performed well during pre-training but remains under post-training development. Google has additionally been pushing its Gemini platform across consumer and enterprise markets, with the Gemini app having surpassed one billion users globally.

For Google, stronger coding performance could help translate that user base into greater adoption among developers and businesses. The company is increasingly competing not only on general-purpose AI models but also on coding assistants, AI agents and enterprise development workflows, where performance, speed and cost are becoming critical factors.