Google launches Gemini for Science with Co-Scientist, Alpha Evolve, and NotebookLM research tools

Gemini for Science is Google's structured pitch to academia and R&D-heavy enterprises: instead of asking researchers to bolt LLMs onto their workflows, package the methodology steps themselves as agents. The Labs collection bundles Co-Scientist (hypothesis generation and literature synthesis), Alpha Evolve (algorithmic discovery), Empirical Research Assistance, and NotebookLM into a single research surface, with corresponding Science Skills landing in Google Antigravity for users who want them in their IDE.

Mechanically, each tool wraps Gemini with domain-specific scaffolding — prompts, evaluators, and tool integrations tuned for scientific workflows. Co-Scientist for example pairs a planning agent with a critic agent so a hypothesis is stress-tested before a researcher invests bench time. Alpha Evolve descends from the DeepMind AlphaEvolve line and applies evolutionary search to algorithms and program synthesis tasks.

Competitive context: this lands the same week Google triggered the $1B Gemini price war and DeepMind expanded its Singapore partnership for pandemic preparedness and healthcare. It also sits opposite Anthropic's Claude Security/Mythos push and OpenAI's job posting for self-training AI — three frontier labs targeting three different verticals for agentic depth (science, security, foundation research). Andrew Ng's new agent course with Google Cloud (image/video generation agents with self-evaluation) reinforces the same agentic-evals theme.

What to watch: adoption signals from named research institutions, whether peer-reviewed papers start citing Co-Scientist or Alpha Evolve in methods sections, and whether OpenAI and Anthropic respond with their own packaged science verticals. The framing matters: Google is positioning Gemini as not just a chatbot but a research collaborator — a much harder claim to dislodge once it's embedded in lab workflows.