What's this about?
Google shared early progress on using Gemini to accelerate scientific discovery in practice — an extension of Co-Scientist, used to collaborate with working scientists across materials science, biology and computer science.
What is Co-Scientist?
A system that treats hypothesis generation as the thing to automate. Rather than answering a question, it proposes candidate hypotheses, criticises them, ranks them and hands a scientist a shortlist worth testing. The bet is that the bottleneck in research is not doing experiments but deciding which experiment is worth doing, which is where a system that can read everything has a real advantage.
Why is it worth noticing?
1,018 likes
- "In the real world, with scientists" is the claim that distinguishes this from the last round of announcements — the previous work was largely retrospective.
- It arrives with OpenAI's Rosalind Workbench and Anthropic's Model Hardware Standard the same week: all three labs are pushing on science simultaneously.
- Read alongside Terminal-Bench-Science and FrontierChallenge from the 2026-08-28 digest — the benchmarks say agents resolve about 30% of real scientific workflows, which is the number to hold against these announcements.