Google's AI research unit is set to release a new model with significantly upgraded coding capabilities as soon as Wednesday, according to the Wall Street Journal.
The model, named 3.8 Flash and known internally as “Skimaki,” has been preferred by Google engineers over Anthropic’s Opus in internal coding tool testing, the report said. Google has reportedly been focusing researchers' time and resources on improving the models' coding capabilities.
Since last year, Google has largely trailed leading AI models from Anthropic and OpenAI. The gap grew larger as the business of AI development shifted toward agentic coding that can create and revise software with less human supervision.
According to The Latent's index measuring the capabilities of current AI frontier models, the top three spots were dominated by Anthropic's Fable and Opus models, followed by OpenAI's GPT-5.6 Sol. Google's Gemini 3.7 Flash ranked 10th out of 16 models.
While placing higher for benchmarks measuring knowledge and reasoning, and preference and communication, the 3.7 Flash was among the bottom ranking models for benchmarks on agentic tool and computer use, and professional and real-world work.
The newest Gemini Flash model's stronger coding performance may bolster Google's competitiveness in AI, but the WSJ report said the company is still unlikely to lead the race as the Flash model is a smaller, cheaper, and faster design with less power compared to bigger models.
The upcoming Flash release comes amid recent structural changes in Google DeepMind, with Co-founder and CEO Demis Hassabis moving into a new chairman role and Koray Kavukcuoglu stepping into Hassabis' role in the position of senior vice president.
Prior to that, Google's AI department saw a wave of departures, including veteran Google executive Jeff Dean, who went on to form his own startup, and prominent researchers Noam Shazeer and John Jumper, who moved to OpenAI and Anthropic, respectively.