The 2026 AI4S boom has turned scientific AI agents from simple assistants into autonomous research partners, and Mira demonstrates this shift by topping multiple benchmarks, cutting task costs to $0.67, and delivering a full‑cycle, secure architecture that real‑world labs can adopt.
The WAIC "Pioneer Youth Talk" wrapped up with high‑density youth talent, policy briefings, and a world‑café format where dozens of young experts dissected self‑improving agents, world models, large‑model limits, multimodal understanding, and AI for science, highlighting both technical insights and emerging risks.
The 11‑billion‑parameter multimodal scientific foundation model “ShenZhen” unifies DNA, RNA, protein, small‑molecule, earth‑system and medical‑image data via native scientific tokens, delivering competitive benchmark results across life, material, earth and medical domains while enabling seamless cross‑modal inference and open community collaboration.
At ICML 2026, ten closed‑door sessions with leading researchers uncovered emerging signals—from next‑generation diffusion language models and data‑centric AI to AI‑driven finance, autonomous agents, AI as an operating system, and AI for science—highlighting the directions that will shape AI research and deployment over the next few years.
The 2015 LINE paper won the WWW 2026 Test of Time award, and its graph‑embedding principles that once powered web search, knowledge graphs, and recommendation systems now underpin GeoFlow models achieving AlphaFold‑level protein structure prediction and breakthrough antibody design, illustrating a ten‑year journey from web graphs to AI‑driven drug discovery.
The article argues that despite rapid advances in AI scientists—automating literature review, hypothesis generation, experimentation, and writing—their impact on scientific speed is limited by a three‑century‑old research protocol, peer‑review bottlenecks, and incentive misalignments, which can only be overcome by redesigning the research artifact itself.
The article argues that despite rapid advances in AI scientists, scientific progress remains limited by the centuries‑old paper format, peer‑review constraints, and incentive structures, and proposes an Agent‑Native Research Artifact to make research forkable and preserve failed experiments, dramatically improving reproducibility and understanding.
On June 30 2026, Anthropic unveiled Claude Science, an AI workbench for scientists, while OpenAI introduced GeneBench‑Pro, a research‑grade benchmark, together highlighting that the next AI battlefield is the laboratory and showcasing early performance gaps between models and human experts.
LabVLA introduces a Vision‑Language‑Action paradigm and a knowledge‑enhanced simulation engine to teach AI systems how to plan and execute real‑world scientific experiments, achieving 71.1%/70.0% success in simulated benchmarks and demonstrating comparable performance on a real Franka robot while highlighting remaining challenges for fully autonomous lab assistants.
LabVLA introduces a vision‑language‑action framework and a knowledge‑enhanced simulation engine to enable AI models to learn and generalize scientific lab manipulation, achieving 71% success on benchmark tasks and demonstrating real‑world performance on a Franka robot, while outlining current limitations and future directions.
Two former Anthropic researchers have founded Mirendil, raising $200 million at a $1 billion valuation to build a recursive self‑improvement AI platform that lets scientists create and iterate their own models, positioning the startup against big‑tech AI moats.
Using OpenAI's Codex, astrophysicist Chi‑kwan Chan generated new coordinate transformations and numerical schemes that could speed up black‑hole plasma simulations by up to a thousandfold, illustrating how AI is moving from answering questions to actively shaping scientific research workflows.
The article analyzes how two Peking University papers presented at ICML 2026 and ACL 2026 introduce BioProBench and BioProAgent to benchmark and enable large language models to safely perform complex wet‑lab experiments, achieving high physical compliance and integrating into a multi‑agent AI4S LAB platform.
The UniCM model unifies ocean‑atmosphere climate modes in a dual‑branch transformer, achieving record‑long ENSO forecasts and revealing emergent predictability across seven key global modes, while offering interpretable attention maps that turn AI from a pure predictor into a climate discovery tool.
The article examines how AI‑native scientific agents demand flexible, secure, and observable infrastructure, and how Alibaba Cloud’s Serverless‑based AgentRun platform delivers extreme elasticity, cost reduction, stateful long‑running support, sandbox security, and full‑chain tracing to enable rapid deployment of tens of thousands of research tools.
Google's Gemini for Science program unifies Gemini, AlphaEvolve, NotebookLM, and Co‑Scientist into a cohesive AI workflow that generates hypotheses, runs computational experiments, and extracts literature insights, aiming to shift scientific bottlenecks from raw compute to intelligent information processing.
A Chinese Academy of Sciences team introduced the MatRIS‑MoE model and the Janus training framework, enabling a 11.5 billion‑parameter universal machine‑learning interatomic potential to be trained on two exascale systems at 1.2 EFLOPS, compressing weeks‑long training into a few hours.
The SimpleTES framework decomposes trial‑and‑error into three scalable dimensions—Concurrency, Length, and Candidates—enabling test‑time scaling that lets open‑source models outperform closed‑source rivals across 21 diverse scientific benchmarks, from LASSO regression to quantum circuit compilation.
A joint Huawei and City University of Hong Kong team combined large language models with evolutionary computation to solve the capacity‑constrained vehicle routing problem, winning the CVRPLib BKS Global Challenge and demonstrating how AI can automate and transform algorithm design, heralding a new paradigm for operations optimization.
This article offers a concise, one‑page overview of more than 100 pivotal AI‑for‑Science papers published in 2025, spanning biomedicine, materials chemistry, climate modeling, astronomy and other fields, and includes key findings, citations, and links for each breakthrough.
An analysis of the 5,984 papers accepted at NeurIPS 2025 shows a decisive move from ever‑larger models toward agents, reasoning‑focused LLMs, efficiency engineering, AI for Science, and trustworthy AI, signaling the transition from a research‑toy era to an engineering‑driven AI ecosystem.
The article traces Demis Hassabis’s journey from chess prodigy to DeepMind CEO, detailing the company’s transition from game‑playing breakthroughs like AlphaGo to scientific initiatives such as AlphaFold and AI4S, while examining ethical debates, Nobel‑prize controversy, and calls for global AI safety standards.
Former OpenAI research VP Liam Fedus and DeepMind veteran Ekin Dogus Cubuk founded Periodic Labs to build an AI‑driven scientific platform that combines autonomous robotic labs, high‑fidelity simulations, and LLM assistants, secured $300 million in seed funding, and assembled a team of over 20 elite researchers to accelerate discovery of room‑temperature superconductors and other materials.
This article explores how AI-driven models and data pipelines are transforming the chemistry and pharmaceutical sectors by accelerating drug design, improving protein‑antibody predictions, automating patent data extraction, and outlining future goals for end‑to‑end AI‑enabled scientific discovery.
In an in‑depth interview, Prof. Hong Liang of Shanghai Jiao Tong University discusses the evolution of AI for Science, the challenges of turning research breakthroughs into real‑world protein‑engineering solutions, the importance of industry‑academia collaboration, and how luck, timing, and focused problem definition drive successful AI adoption.
The article discusses how AI for Science, exemplified by a 4.5‑billion‑parameter weather model and growing international initiatives, is reshaping scientific research, fostering interdisciplinary collaboration, and driving policy and institutional investments to accelerate innovation across domains.
This article reviews the role of graph neural networks in AI for science, covering background, the evolution of GNN models, applications in physics and biomedicine, recent advances in Euclidean equivariant GNNs, and the authors' own contributions such as GMN and GROVER, concluding with key distinctions between traditional GNNs and science‑focused approaches.
