AI for Science
Scientific discovery, drug design, materials science, and AI-driven research acceleration. Accelerating humanity's understanding of the world.
Key Benchmarks
Recent Papers
TerraNova: A Foundation Model for the Anthropocene
Carlos Rodriguez-Pardo, Massimo Tavoni
MOT-SR: Multi-Objective Tool-Augmented Scientific Equation Discovery with Large Language Models
Boxiao Wang, Runxiang Wang, Kai Li +4 more
ProofCouncil: An LLM Agent for Solving Open Mathematical Problems
Johannes Schmitt, Tim Gehrunger, Jasper Dekoninck +4 more
LongMedBench: Benchmarking Medical Agents for Long-Horizon Clinical Decision-Making
Yanzhen Chen, Zihan Xu, Xiaocheng Zhang +4 more
Toward Auditable AI Scientists: A Hypothesis Evolution Protocol for LLM Agents
Izumi Takahara, Teruyasu Mizoguchi
Discrete Diffusion Language Models for Interactive Radiology Report Drafting
Max Van Puyvelde, Halil Ibrahim Gulluk, Wim Van Criekinge +1 more
PairSAE: Mechanistic Interpretability from Pair Representations in Protein Co-Folding
Giosue Migliorini, Aristofanis Rontogiannis, Grigori Guitchounts +3 more
Multi-Objective Molecular Generation with Frequency-Controlled Evolutionary Dynamics
Elia Colleoni, Paolo Guida, Didier Barradas-Bautista +1 more
A Global Context-Aware Deep Neural Network for Enhanced Brain Tumor Segmentation using Magnetic Resonance Images
Sourjya Mukherjee, Ananya Bhattacharjee, R. Murugan
Frontier LLM-based Agents Can Overcome the Ontology Curation Bottleneck for Natural Phenotypes
James P. Balhoff, Hilmar Lapp
Recent Milestones
US launches $5B Genesis AI-for-science push
On July 22, 2026, the White House announced more than $5 billion in federal commitments for the Genesis Mission, a national effort to apply AI and advanced computing to scientific discovery. Over 15 U.S. agencies will fund hundreds of projects across health, energy, infrastructure, and manufacturing using shared AI data and compute platforms.
Genesis Mission: $5B+ bet on AI for US science
On July 22, 2026, the White House announced more than $5 billion in federal commitments to expand the Genesis Mission, a national AI-for-science initiative anchored at the Department of Energy. Microsoft, Google, and OpenAI simultaneously unveiled new in‑kind AI and cloud commitments, while NSF, NIH and leading universities detailed funded research campaigns under the program.
CuspAI raises $450M for AI materials foundry
Cambridge-based CuspAI closed a $450 million Series B round at a $2.6 billion valuation, announced on July 20, 2026. The funding, co-led by Kleiner Perkins and NEA with participation from Bezos Expeditions and the UK’s Sovereign AI Fund, will expand its generative AI platform for discovering new materials and launch an AI Materials Foundry network.
Claude Science Makes AI a Lab Co‑Researcher
Anthropic has launched Claude Science, an AI workbench that unifies scientific databases, tools and compute into a single environment for researchers. The product, announced June 30 and covered widely on July 1, 2026, runs on existing Claude models such as Opus 4.8 and includes features like multi-agent workflows, citation checking and grants of up to $30,000 for selected projects.([anthropic.com](https://www.anthropic.com/news/claude-science-ai-workbench?id=Claude4&utm_source=openai))
Claude helps prove decade‑old jamming conjecture
On July 1, 2026 Phys.org reported that physicists Giorgio Parisi and Francesco Zamponi used Anthropic’s Claude model as a collaborator to help prove a 10‑year‑old conjecture in jamming theory. Claude reproduced prior numerical work and suggested a key idea that the researchers refined into a full mathematical proof.
AutoLabs Agents Drive 5–10x Faster Robot Science
On June 25, 2026, the U.S. Department of Energy’s Pacific Northwest National Laboratory unveiled AutoLabs, an agentic AI system that translates scientists’ experimental goals into detailed instructions for an autonomous lab robot. PNNL says AutoLabs can boost experimental throughput by 5–10x and has been open-sourced alongside a Scientific Reports paper.
AI Finds New Antibiotics for Drug‑Resistant Gonorrhoea
On June 18, 2026, Open Access Government reported that researchers at Harvard’s Wyss Institute, MIT and the Broad Institute used a deep learning workflow to screen ~6 million compounds and identify new candidates effective against multidrug‑resistant Neisseria gonorrhoeae. The work, detailed in Science Translational Medicine and a Wyss Institute release from June 17, shows AI‑discovered compounds targeting previously underused bacterial pathways with promising in vitro and in vivo results. ([openaccessgovernment.org](https://www.openaccessgovernment.org/machine-learning-identifies-new-compounds-to-target-antibiotic-resistant-gonorrhoea/210692/))
Bezos’ Prometheus: $41B bet on ‘General Engineer’ AI
In a June 11, 2026 Financial Times interview republished by Chile’s Diario Financiero, Jeff Bezos said AI will usher in multiple “golden ages” rather than mass job destruction. He outlined his vision for Prometheus, a new AI lab valued at about $41 billion that aims to apply AI to manufacturing and engineering.([df.cl](https://www.df.cl/internacional/ft/jeff-bezos-asegura-que-la-ia-traera-edades-doradas-y-no-una-destruccion))
AI designs first universal coronavirus vaccine
A Cambridge-led team has completed the first human trial of a ‘universal’ coronavirus vaccine whose key antigen was designed entirely by artificial intelligence, with early results published June 5, 2026. Coverage on June 5–6 reports the DNA vaccine was safe in 39 volunteers and generated modest but promising immune responses against the Sarbeco coronavirus family. ([sciencedaily.com](https://www.sciencedaily.com/releases/2026/06/260605023357.htm?utm_source=openai))
OpenAI Launches Rosalind Biodefense With GPT-Rosalind
On May 29, 2026, OpenAI launched the Rosalind Biodefense program to sponsor vetted developers building biodefense and pandemic preparedness tools with its life-sciences model GPT‑Rosalind. The company also began extending trusted access to GPT‑Rosalind for select U.S. government and allied public‑health and biodefense agencies, after briefing the White House and several federal bodies on the effort.([openai.com](https://openai.com/index/strengthening-societal-resilience-with-rosalind-biodefense/))
AlphaProof Nexus cracks open Erdős problems
Google DeepMind said on May 26, 2026 that its AlphaProof Nexus system has formally solved nine open Erdős problems and 44 OEIS conjectures using Lean-verified proofs. CEO Demis Hassabis stressed the system is “still not AGI” but positions it as a new tool for rigorous mathematical research.
GalaxyVS Screens 100B Molecules in Seconds
Chinese media report that researchers at the National Supercomputing Center in Tianjin and Tsinghua University have deployed "GalaxyVS", an AI-driven virtual screening platform built on the DrugCLIP model. According to MyDrivers, the system can perform sub-second retrieval over a 100‑billion–molecule library by mapping protein pockets and ligands into a shared vector space on Tianhe supercomputers.([toolify.ai](https://www.toolify.ai/daily-ai-news))
Robin AI Scientist Automates Months of Lab Work
Nonprofit FutureHouse has published a Nature paper unveiling "Robin", a multi-agent AI system that autonomously generates hypotheses, designs experiments and analyzes wet-lab data. An AIbase report on May 25, 2026 highlights that Robin completed roughly four months (around 900 hours) of human research work in about two hours and identified a candidate treatment for dry age-related macular degeneration.([toolify.ai](https://www.toolify.ai/daily-ai-news))
Anthropic’s $400M Bet on AI Drug Discovery
Anthropic has acquired stealth AI biotech startup Coefficient Bio in an all‑stock deal worth roughly $400 million, pulling a sub‑10‑person team of ex‑Genentech researchers into its healthcare and life sciences unit in early April 2026. The deal, reported by multiple outlets and confirmed by sources close to Anthropic, aims to apply Claude‑style foundation models to end‑to‑end drug discovery workflows.([winbuzzer.com](https://winbuzzer.com/2026/04/05/anthropic-acquires-coefficient-bio-400m-ai-pharma-startup-xcxwbn/))
Anthropic bets $400M on AI drug discovery
On April 4, 2026, multiple outlets reported that Anthropic has acquired New York–based AI biotech startup Coefficient Bio in an all‑stock deal valued at about $400 million. The roughly nine‑person team will join Anthropic’s life sciences unit to build domain‑specific models for drug discovery and biomedical R&D.
Anthropic Buys Coefficient Bio for $400M
Anthropic has acquired New York–based biotech startup Coefficient Bio in a deal valued at about $400 million, mostly in stock. The move, reported on April 3, 2026, folds Coefficient Bio’s biology-focused AI team into Anthropic’s Health Care Life Sciences group to apply Claude models to drug R&D and clinical strategy.
Insilico Opens ‘Science Gym’ for Frontier LLMs
On January 22, 2026, Insilico Medicine announced a new ‘Science MMAI gym’ service designed to train general‑purpose LLMs such as GPT and Qwen to perform better on biology and chemistry tasks. The Hong Kong‑listed biotech says its pipeline can boost model performance by up to 10x on key scientific benchmarks using domain datasets, reward models and reinforcement learning.
GPT‑5.2 helps solve long‑standing Erdős problem
On January 18, 2026, Eclipse founder Neel Somani said he used OpenAI’s GPT‑5.2 Pro to solve Erdős Problem #281, a number theory problem open since 1980. Fields Medalist Terence Tao and other mathematicians reviewed and accepted the proof, and OpenAI president Greg Brockman highlighted the result as a major milestone for AI-assisted science.
Merge Labs raises $250M for brain–AI links
Brain‑computer interface startup Merge Labs emerged from stealth with a reported $252 million seed round, led by OpenAI and other investors. Coverage on January 16 highlights the lab’s plan to build high‑bandwidth, largely noninvasive BCIs that integrate tightly with advanced AI systems.
KAIST AI maps B‑ and T‑cell targets for cancer vaccines
A joint team from KAIST and biotech firm Neogenlogic announced an AI model that predicts which tumor neoantigens will trigger robust B-cell and T-cell responses, enabling personalized cancer vaccine design. The model, validated on large genomic and clinical datasets, is described as the first AI framework to jointly model B-cell immunogenicity and T-cell responses, with clinical trials targeted for 2027.