Google's Foundation Model Third Place Crisis
When Google kicks off its annual developer conference, I/O, tomorrow in Mountain View, California, it will do so under a shadow it hasn't faced in years: the company is now a clear third place in the foundation model race. A year ago, at I/O 2025, Google was riding high on Gemini 2.5 Pro, and distinguishing among top-tier large language models felt like a subjective splitting of hairs. That landscape has shifted dramatically, driven primarily by one metric: coding capabilities.
Over the past six months, Anthropic's Claude Code and OpenAI's Codex have established a decisive lead in AI-assisted software development. According to MIT Technology Review, these systems are so dramatically superior to Google's own offerings that the company has reportedly had to allow some engineers at DeepMind to use Claude for their work — a concession that underscores the depth of the crisis.
The stakes are high because coding benchmarks have become the de facto standard for measuring foundation model intelligence. A model that excels at code generation, debugging, and agentic workflows is perceived to be more capable across all domains. Google's fall from parity to third place represents not just a competitive setback but a reputational one, particularly within the developer community that I/O is meant to court.
DeepMind's Secret Coding Offensive
Behind the scenes, Google is taking this challenge seriously. According to reporting from The Information, a new AI coding team has been formed within DeepMind, the company's elite AI research division. More tellingly, the Los Angeles Times has reported that John Jumper — who shared a 2024 Nobel Prize in chemistry with DeepMind CEO Demis Hassabis for their work on AlphaFold — is lending his talents to the coding effort.
Jumper's involvement signals that Google views this as a research problem worthy of its best minds, not just a product iteration. At I/O, the company is expected to release a major new coding product, likely an update to its Antigravity agentic coding platform. Antigravity, first previewed at last year's I/O, was designed to let developers describe software features in natural language and have the AI autonomously build, test, and deploy them.
However, we shouldn't expect a revolution overnight. Googlers themselves have had access to internal models and products that are substantially ahead of those released to the public, yet they were still reportedly fighting over who got access to Claude Code last month. Unless the company has made astonishing progress in the intervening weeks, Google probably won't leapfrog back to the coding frontier in the next two days. The better bet is a credible step forward that narrows the gap — enough to keep enterprise customers from defecting to Anthropic or OpenAI.
Science Remains Google's Conspicuous Strength
While coding may be Google DeepMind's weakness, science is its conspicuous strength. It remains the only frontier AI company to have earned a Nobel Prize. As large language models have come to dominate the AI-for-science landscape, Google has only solidified its lead. Last year, the company released multiple scientific AI tools, including the AI co-scientist, which formulates hypotheses and research plans in response to user questions — described as an "oracle" by one Stanford scientist — and AlphaEvolve, a system that iteratively discovers new solutions for mathematical and computational problems.
Any new scientific tools announced at I/O will be particularly noteworthy because they underscore a strategic differentiator that OpenAI and Anthropic cannot easily replicate. DeepMind's culture, born from academic neuroscience and physics, gives it an institutional advantage in fields like protein folding, drug discovery, and materials science. For enterprise customers in biotech, pharma, and engineering, Google's scientific AI capabilities may outweigh its coding shortfalls.
I will also be watching closely for moves in health and medicine. Google has been doing some of the best research on LLM-based health tools, but OpenAI has defined the health AI conversation since the release of ChatGPT Health in January. Google has announced it will make its AI-powered Health Coach publicly available tomorrow, but promotional material suggests the tool is geared more toward fitness and diet advice than medical concerns. This raises an open question: Is Google exercising appropriate caution in a high-stakes domain, or has it fallen behind here too?
The Drama-Free Developer Conference?
While Google fans congregate in Mountain View, roughly 30 miles north in Oakland, the Elon Musk v. Sam Altman trial is wrapping up. The past few months have seen more than their fair share of AI CEO drama — before the trial, the animosity between Altman and Anthropic CEO Dario Amodei took center stage as both companies negotiated deals with the US Department of Defense. But DeepMind's Hassabis has, for the most part, steered clear of such drama. He effectively presents himself as a Nobel Prize-winning nerd, and if he has written screeds about any of his peers, they haven't been leaked to the press.
That doesn't mean Google is controversy-free. Last month, a group of 600 employees, many from DeepMind, sent a letter to CEO Sundar Pichai protesting an impending DoD deal. Google signed that deal the next day. Hassabis, Pichai, and other executives will surely do their best to skirt such touchy subjects while on stage, but controversies have a way of worming into Q&A sessions and hallway conversations. How Google handles these internal tensions — especially the moral qualms of its AI researchers — may be just as consequential as any product launch.
What's at Stake for Developers and the Industry
For developers attending I/O, the coding update is the headline. Google's ecosystem — Android, Firebase, Cloud — is deeply intertwined with developer tools. If the company can ship a coding assistant that approaches Claude Code in quality, it could slow the erosion of its developer mindshare. But if the update underwhelms, the narrative of Google as an AI laggard will harden.
There's also a broader industry implication: The coding arms race is forcing every AI lab to invest heavily in agentic capabilities. Codex and Claude Code can now handle multi-file refactoring, test generation, and even pull request reviews with minimal human intervention. Google's Antigravity update will need to demonstrate similar autonomy to be taken seriously.
Forward-looking, I/O 2026 may be remembered as the moment Google acknowledged it can't win on every front — so it must double down on the ones where it has unique advantages, like science and health, while rebuilding its coding competence from within. The next two days will reveal whether that strategy is viable or merely a holding action.
Commentaires