Google's DeepMind plans to introduce a new model, Gemini 3.8 Flash, with significantly enhanced programming capabilities. Internally codenamed "Skimaki," it is expected to launch as early as September 2 (local time). In internal testing, engineers showed a preference for this model over Anthropic's Opus, significantly narrowing the gap in programming capabilities between Google and its competitors, Anthropic and OpenAI. This launch comes amid personnel adjustments at Google DeepMind, with the new leader emphasizing the need to accelerate execution. Previously, Google faced setbacks in the model development race, with the Gemini series lagging in key performance areas such as programming. Gemini 3.8 Flash belongs to the Flash series, designed to be smaller, cheaper, and faster to run, though its performance does not match that of the flagship Pro series models. Google has dedicated more research time and computational power to reinforcement learning to enhance the model's programming capabilities. Additionally, Google has recruited Barret Zoph, former head of training at OpenAI, to focus on reinforcement learning and post-training areas.