---
format: "aidr-story-markdown/v1"
id: "420880e0ff8fafb5db284ee8f4447786eb9b2a9a4e8e815ef3f65c9768d6bf43"
canonical_url: "https://aidr.today/420880e0?lang=en"
title: "Google DeepMind Co-Scientist Grows 3 Semiconductors on First Attempt in Real World Debut"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-08-29T14:38:30.000Z"
category: "Research"
topics: ["google","gemini","agent","reasoning","benchmark"]
source_urls: ["https://huggingnews.com/ai/google-deepmind-co-scientist-grows-3-semiconductors-on-first-attempt-in-f5be6879","https://x.com/timstro/status/2093426468756107722","https://x.com/_philschmid/status/2093694057818009934","https://x.com/deployedmind/status/2093656894095167886"]
summary: "An extension of the Gemini model designed a safe precursor route for 2D MXene synthesis and predicted the swarming morphology of engineered E. coli. Developed by Google DeepMind, Co-Scientist proposed lab recipes that produced 3 atom thin semiconductors on the first attempt and created an inference architecture that outperformed six frontier models on the HealthBench Hard and Professional benchmarks. The tool transitions AI agents from simulations to closed-loop science by interfacing with physical hardware. In tests of AI-generated research, the system reduced serious fabricated results in papers from 90% to 4% and validated its biological predictions against unpublished wet-lab morphological measurements."
---

# Google DeepMind Co\-Scientist Grows 3 Semiconductors on First Attempt in Real World Debut

> [Open the canonical story](<https://aidr.today/420880e0?lang=en>)

**Published:** 2026-08-29T14:38:30.000Z
**Category:** Research
**Topics:** google, gemini, agent, reasoning, benchmark

## Summary

An extension of the Gemini model designed a safe precursor route for 2D MXene synthesis and predicted the swarming morphology of engineered E\. coli\. Developed by Google DeepMind, Co\-Scientist proposed lab recipes that produced 3 atom thin semiconductors on the first attempt and created an inference architecture that outperformed six frontier models on the HealthBench Hard and Professional benchmarks\. The tool transitions AI agents from simulations to closed\-loop science by interfacing with physical hardware\. In tests of AI\-generated research, the system reduced serious fabricated results in papers from 90% to 4% and validated its biological predictions against unpublished wet\-lab morphological measurements\.

## Sources

- [Story source](<https://huggingnews.com/ai/google-deepmind-co-scientist-grows-3-semiconductors-on-first-attempt-in-f5be6879>)
- [Story source](<https://x.com/timstro/status/2093426468756107722>)
- [Supporting source](<https://x.com/_philschmid/status/2093694057818009934>)
- [Supporting source](<https://x.com/deployedmind/status/2093656894095167886>)