Google Coding AI Push Puts Pressure On Anthropic

Google is reportedly preparing a stronger coding AI model, a sign that the company is not willing to let Anthropic and OpenAI own the developer market uncontested.

The Wall Street Journal says the model, internally associated with names such as Skimaki and 3.8 Flash, has narrowed the coding gap with Anthropic. Google has not publicly announced that specific model, so the sensible way to read this is as a report about where the Gemini roadmap appears to be heading.

The direction itself is not surprising. Google has been pushing developers toward the Gemini API, Gemini Code Assist and other tools that make its models useful inside software work. The company knows that coding is one of the clearest places where AI can turn into real productivity and real revenue.

Coding models matter because developers are not only asking for answers. They are asking AI systems to read codebases, fix bugs, write tests, generate interfaces, review pull requests and sometimes operate as agents. If a model performs well there, it can become part of daily work very quickly.

Anthropic has enjoyed strong momentum because Claude is widely liked by developers for code reasoning, editing and long-context work. OpenAI is still powerful in the category, especially through ChatGPT and agentic coding tools. Google has enormous infrastructure and research depth, but it has had to prove that Gemini can feel as reliable and useful in real development workflows.

The competition is not only American. Chinese AI labs are also putting pressure on pricing and performance. We recently wrote about Z.ai and China AI price pressure, where strong open and low-cost models are forcing the whole market to explain why users should pay more.

That is why Google’s coding push is strategically important. The company owns Android, Chrome, Google Cloud, Workspace and a large developer ecosystem. A top-tier coding model can connect all of those pieces, from app development to cloud deployment and enterprise automation.

The risk is that developers are quick to switch. If a model is slow, expensive, unreliable or poor at handling real repositories, loyalty disappears. Benchmarks matter, but developer trust matters more. A model that looks good in a leaderboard still has to survive messy production code.

The bigger story is that AI coding is becoming the front door to agentic work. Once a model can understand and modify software reliably, it can also become the base for workflow automation, debugging, security reviews and internal tools. That makes the coding race one of the most important business fights in AI.

If Google can close the gap, it will put new pressure on Anthropic, OpenAI and Chinese challengers at the same time. If it cannot, it risks watching the developer workflow shift toward rivals even while Google owns some of the world’s most important computing platforms.