Google's Coding-First AI Model Push Comes Amid DeepMind Restructuring and Gemini Delays

What happened

Google is reportedly on the verge of shipping a coding-first AI model, according to Gizmodo. The report lands amid a broader upheaval inside Google's AI division, where DeepMind has lost its independence, its founder has stepped back from the CEO role, and the next Gemini flagship has slipped behind rivals on coding benchmarks.

The troubles trace back months. In July, Bloomberg reported—citing ten current and former Google employees—that the next Gemini Pro upgrade was months behind schedule because its coding capabilities fell short of internal goals. Google had been expected to release the upgrade at its May developer conference but could not close the gap with Anthropic and OpenAI, both of which had released models outperforming Gemini in writing code. Alphabet shares dropped more than three percent on that news. Google reportedly attempted to update Gemini's training data to improve coding abilities, but results were disappointing.

By early August, the restructuring was underway. On August 5–6, Demis Hassabis stepped back from CEO to chairman, and AI researcher Koray Kavukcuoglu took over day-to-day operations—without the CEO title. According to The Decoder, citing The Guardian and Pathfounders, a former Google manager said "the era of DeepMind as an independent actor is over," and a current employee called the lab "just another subdivision" of Google. Sergey Brin is reportedly taking a more influential role, pressing for faster commercialization.

Two of Gemini's original technical co-leads, Jeff Dean and Oriol Vinyals, left Google and founded a startup together, removing senior engineering leadership from the project. Engineers have also been departing for Anthropic specifically, according to Bloomberg's sources. Employees cited long-running frictions over multiple project leads, slow approval processes, and tight TPU and cloud compute supply as reasons for slowed development.

Now, Gizmodo reports Google is preparing to ship a coding-first model—a targeted response to the area where Gemini has struggled most.

Why it matters

Coding has become the defining battleground for frontier AI models. Developers are the highest-value, most vocal user base for AI tools, and the vendor that wins their loyalty gains durable platform leverage. Google's inability to match Anthropic's Claude or OpenAI's models on coding tasks is not a niche weakness—it's a signal that the company's model development pipeline is stuttering.

The DeepMind restructuring compounds the problem. Losing Hassabis as CEO, losing Dean and Vinyals as technical leaders, and folding DeepMind's research autonomy into Google's product org all suggest a company in catch-up mode. Whether a coding-first model can close the gap remains an open question, but the pivot signals Google recognizes where it's falling behind.

What to watch

• Official announcement details: No model name, pricing, API access, or release date has been confirmed. Watch Google Cloud and Vertex AI channels for an official launch.
• Benchmark performance: Whether the coding-first model can match or beat Claude or GPT on established coding benchmarks will determine its credibility.
• Gemini Pro timeline: The flagship upgrade is reportedly delayed roughly two months. Watch for whether it ships with improved coding capabilities or remains behind.
• Talent retention: Further departures to Anthropic or other rivals would deepen concerns about Google's ability to execute.