One year is all it took for Google I/O to flip from a victory lap for Gemini 2.5 Pro into a credibility test for Google’s entire AI strategy.
When Google opens its annual developer conference in Mountain View, California, it will arrive as “a clear third place in the foundation model race,” according to MIT Technology Review. That framing matters because Google is not a research lightweight trying to break in. It is the company behind DeepMind, Gemini, and Nobel-winning AI work. The problem is sharper: Google has world-class AI research, but the market narrative has shifted toward tools that developers actually use every day.
That makes this I/O less about spectacle and more about proof. Google needs to show that Gemini can become a serious working layer for coding, science, health, and consumer products — not just a model family with impressive demos.
One Year After Gemini 2.5 Pro, Google Is Fighting the “Third Place” Label
At Google I/O 2025, the company was still riding momentum from the March launch of Gemini 2.5 Pro. MIT Technology Review describes that period as one when distinguishing among top-tier large language models “often felt like a subjective splitting of hairs.”
That is no longer the story.
The reputational center of the foundation model race has moved toward coding. On that front, Google’s tools are being judged against Anthropic’s Claude Code and OpenAI’s Codex. MIT Technology Review says those systems have outperformed Google’s offerings so clearly that Google has reportedly had to allow some DeepMind engineers to use Claude for their work.
Google’s problem is not that it lacks AI talent. It is that, in the most visible developer workflow of the moment, rivals are setting the pace.
MLXIO analysis: This is the core tension beneath I/O. Google can still announce powerful models, but the credibility test is narrower and harsher: can Gemini win back developers who now associate frontier coding agents with Anthropic and OpenAI?
That is why any update to Antigravity, Google’s agentic coding platform, would matter. MIT Technology Review says a major coding release would not be surprising. But the same source cautions that expectations should stay grounded. If Google employees with access to stronger internal systems were still reportedly competing for access to Claude Code last month, a two-day conference is unlikely to erase the gap.
Two Coding Rivals Now Define Google’s AI Problem
The most damaging comparison for Google is not abstract model quality. It is workflow ownership.
| Area | Google’s position from the source | Rival pressure named in the source |
|---|---|---|
| Foundation models | Described as “a clear third place” going into I/O | OpenAI and Anthropic lead the narrative |
| Coding tools | Google’s offerings have been “outgunned” | Claude Code and Codex |
| Science AI | A visible strength for Google DeepMind | No equivalent rival claim supplied |
| Health AI | Strong research, but public product posture appears cautious | ChatGPT Health has defined the conversation since January |
This table shows why raw model capability is no longer enough. Google has to compete where users feel the difference: code generation, debugging, agent reliability, model access, and product packaging.
The source does not provide Gemini usage, API adoption, Cloud AI revenue, Workspace subscriptions, or Search AI rollout figures. That absence is itself useful. If Google wants to shift the narrative at I/O, it will need more than stage demos. Public adoption signals would do more to weaken the “third place” claim than another polished benchmark slide.
For readers tracking Google’s broader AI positioning, MLXIO’s prior coverage of Google Sparks AI Race with Gemini 3.5 Flash’s Breakthrough Speed and Singularity Bet Recasts Google I/O's AI-Driven Science offers useful context on how much pressure the company faces to connect model capability with visible product momentum.
DeepMind’s Strongest I/O Card Is Science, Not Code
Google’s best counter-narrative is not in coding. It is in AI for science.
MIT Technology Review points out that Google DeepMind is the only frontier AI company to have earned a Nobel Prize. John Jumper shared a 2024 Nobel Prize in chemistry with DeepMind CEO Demis Hassabis for work on AlphaFold, the protein structure prediction software. Jumper is also reportedly lending his talents to Google’s AI coding push, according to the source’s summary of reporting from the Los Angeles Times.
That detail cuts two ways. It shows Google is taking the coding problem seriously enough to pull in elite scientific AI talent. It also highlights where DeepMind’s reputation is strongest: not in coding assistants, but in research systems that solve hard scientific and computational problems.
Last year, Google released multiple scientific AI tools, including:
- AI co-scientist: A system that formulates hypotheses and research plans in response to user questions.
- AlphaEvolve: A system that iteratively discovers new solutions for mathematical and computational problems.
The AI co-scientist has been described as an “oracle” by one Stanford scientist, according to MIT Technology Review’s cited source.
MLXIO analysis: If Google announces new scientific tools at I/O, they may not dominate the consumer AI headlines. But they could be more strategically durable than another chatbot feature. Science is one area where Google can still argue from achievement, not aspiration.
Health Coach Shows Google’s Caution Problem — Or Its Discipline
Health is the more ambiguous test.
MIT Technology Review says Google is doing some of the best research on LLM-based health tools, but OpenAI has shaped the health AI conversation since the release of ChatGPT Health in January. Google has announced that its AI-powered Health Coach will become publicly available tomorrow.
The key detail is positioning. Promotional material suggests Health Coach is focused more on advice around fitness and diet than on direct medical concerns.
That leaves two possible readings.
- Caution reading: Google is moving deliberately because health advice is a high-stakes domain where errors can carry real harm.
- Lagging reading: Google has strong research but is again slower to package it into a product that captures public attention.
The source does not settle that question. I/O may clarify whether Health Coach is a narrow wellness product, a first step toward broader medical AI, or a deliberately constrained consumer feature.
600 Employees, One DoD Deal, and a Trial 30 Miles Away
Google’s I/O stage will not exist in a vacuum.
While Google fans gather in Mountain View, the Elon Musk v. Sam Altman trial will be wrapping up roughly 30 miles north in Oakland, according to MIT Technology Review. The source frames this as part of a wider stretch of AI CEO drama, including tensions between Sam Altman and Anthropic CEO Dario Amodei as Anthropic and OpenAI worked to negotiate deals with the US Department of Defense.
DeepMind’s Demis Hassabis has mostly avoided that kind of public conflict. MIT Technology Review describes him as presenting himself as a Nobel Prize-winning nerd rather than an executive trading public blows.
Google still has its own controversy. Last month, 600 employees, many of them working for DeepMind, sent a letter to CEO Sundar Pichai protesting an impending DoD deal. Google signed that deal the next day.
MLXIO analysis: This is where Google’s neutral, research-first image gets stress-tested. The company can try to keep I/O focused on products, but AI infrastructure, defense work, and employee dissent are now part of the industry’s operating reality. If those tensions surface, they could complicate Google’s preferred message: that it is the responsible adult in the AI race.
After I/O, Google’s Comeback Will Be Measured in Use, Not Applause
The most likely I/O storyline is clear: Google will try to show progress in coding, strength in science, and caution in health.
The harder question is whether any of it changes behavior.
A credible Google AI reset would need evidence after the conference, not just during it. The strongest signals would include visible developer adoption of any new coding release, broader use of Gemini tools, clear uptake of Health Coach, and serious attention around any new AI-for-science systems. The weakest signal would be a familiar one: impressive demos followed by continued reliance on rival tools where the work is hardest.
The thesis to watch is simple. If Google’s announcements make Gemini feel like a daily tool for developers and researchers, I/O could mark the start of a real comeback. If the company mainly shows catch-up features while Claude Code and Codex remain the reference points, the “third place” label will survive the week.
The Bottom Line
- Google I/O is becoming a credibility test for whether Gemini can compete in practical AI workflows.
- Developer adoption now appears central to the foundation model race, especially in coding tools.
- Google’s challenge is turning world-class AI research into products that users and engineers rely on daily.
