Google prepares Gemini 3.8 Flash to narrow AI coding gap, WSJ reports

Alphabet Inc.'s Google DeepMind is set to unveil an artificial intelligence model, Gemini 3.8 Flash, potentially as early as Wednesday, according to The Wall Street Journal. This new model aims to address a weakness in coding capabilities that Google has faced against competitors OpenAI and Anthropic. Known internally as "Skimaki," Gemini 3.8 Flash is part of a strategic push to enhance performance in the enterprise market, where coding has become a crucial application for generative AI.

In internal testing on Google's Jetski platform, engineers found Gemini 3.8 Flash outperformed Anthropic's Claude Opus model. This advancement aligns with Google's priority to leverage agentic coding, a major business use case for AI technology. The rollout could stabilize market sentiment following recent leadership changes, with new CEO Koray Kavukcuoglu emphasizing a faster pace of development post a leadership transition.

Despite its smaller architecture, Gemini 3.8 Flash offers lower compute requirements for modifications, enabling parallel research efforts through reinforcement learning. This model is expected to be more agile than Google's flagship Pro series, which suffered delays due to the resource-intensive nature of its updates. DeepMind is simultaneously working on the upcoming Gemini 4 model, which is in posttraining stage following strong pretraining evaluations.

The market reaction to Google's announcement is evident, with Alphabet shares rising 0.6% in after-hours trading as investors consider the strategic impact of the upcoming release.