Google announced Gemini for Science at I/O, uniting its LLM-based research systems including AI Co-Scientist and AlphaEvolve and offering researchers a route to apply for access.
The package frames a strategic tilt toward agentic, LLM-driven research that can generate hypotheses and optimise algorithms, and Demis Hassabis, CEO of Google DeepMind, told the I/O audience he believes we are “standing in the foothills of the singularity.”
Google is not abandoning specialised models: AlphaGenome and AlphaEarth Foundations were released last summer and the newest version of WeatherNext shipped in November.
Specialised tools remain widely used, the company says, with AlphaFold protein-structure predictions used by over three million researchers and Isomorphic Labs raising a $2 billion Series B to apply those technologies to drug discovery.
The personnel and priority signals are visible: the Los Angeles Times reported that Google fellow John Jumper has moved from science-specific work to AI coding. That's a shift the source links to a broader emphasis on building coding and agentic capabilities.
Related reading
- Google Cloud launches Gemini Enterprise Agent Platform as centrepiece of Cloud Next overhaul
- Apple's decision to open Siri to rival AI models may be its smartest move, and its biggest concession
- Adobe deepens NVIDIA and WPP ties to embed autonomous AI agents in marketing
Agentic systems are already making contributions outside Google; OpenAI announced that one of its models disproved an important mathematics conjecture using a general-purpose reasoning model.
Google presents Gemini for Science as an accelerant for human researchers rather than a replacement, and Hassabis frames the next decade as one for using AI as “an amazing tool to help scientists.”
