explainx / blog
Claude Science is Anthropic's AI workbench for researchers, integrating PubMed, Jupyter, R, HPC, and 60+ domain skills into one auditable scientific environment.
Jun 25, 2026
Claude Tag puts Claude inside Slack as a persistent, proactive team member. The @ClaudeDevs team shared how they use it to write 65% of their own code — including most of what built Claude Tag itself. Here are 15 specific use cases across engineering, ops, data, support, and documentation that show exactly what delegating to @Claude looks like in practice.
Jun 24, 2026
Claude Tag is Anthropic's evolution of Claude Code — now multiplayer, proactive, and woven into Slack. Tag @Claude in any channel and it breaks tasks into stages, works through them with the tools it has, and responds in the thread. One Claude per channel means teammates pick up exactly where you left off. It learns from channel history, takes initiative in ambient mode, and can plan work over hours or days. 65% of Anthropic's own product team's code now comes from their internal version.
Jun 19, 2026
Claude Code Artifacts lets teams deploy shareable HTML apps from inside a coding session—private by default, shared within your org. Here's how it compares to Lovable, v0, Bolt, and Codex Sites.
TL;DR: Anthropic launched Claude Science on June 30, 2026 — a dedicated AI workbench that brings together PubMed, Jupyter, R, cluster terminals, and on-demand GPU compute inside a single environment built for researchers. The beta is available now for Claude Pro, Max, Team, and Enterprise users at claude.com/science. Early adopters at Manifold Bio, the Allen Institute, and UCSF are already reporting order-of-magnitude acceleration on genomics, drug target nomination, and long-form literature review.
AI assistants are useful, but a general-purpose chat interface is not how scientists actually work. A real research workflow spans literature search, data wrangling, statistical modeling, figure generation, code review, and citation checking — often across half a dozen tools that do not talk to each other. Switching context between a terminal, a Jupyter notebook, a reference manager, and a chat window introduces friction that slows every step.
Claude Science is Anthropic's answer to that problem. Rather than asking researchers to copy-paste results between tools, the workbench connects Claude directly to the instruments scientists already depend on and keeps every decision — the code that produced a figure, the full conversation that led to it — in one auditable record. The result is a research environment that accelerates work without hiding it.
This launch builds on the biological research capabilities first demonstrated in the Claude Fable 5 and Mythos 5 release, where Mythos-class models showed autonomous drug design and single-cell genomics at speeds that rivaled or exceeded skilled human scientists.
At its core, Claude Science is a desktop application — available on macOS and Linux — that connects Claude to a configurable set of research tools through a persistent project workspace. Think of it as a research environment that happens to have a frontier AI model as the primary interface.
When you open a Claude Science project, you are not just starting a conversation. You are opening a workspace that remembers your data sources, your analysis history, your installed skills, and every artifact the session has produced. Each project carries its full message history alongside the reproducible figures and code that made them, so anyone on your team — or a future reviewer — can trace exactly how a result was generated.
Compute runs wherever you need it. Claude Science can execute code on your local laptop, on a Linux workstation, on an HPC login node via SSH, or on Modal for GPU-intensive jobs like protein structure prediction or large-scale embedding. Crucially, your datasets never leave your own infrastructure. Only the context window is transmitted to Anthropic.
Every figure, table, or statistical result generated inside Claude Science is paired with the code that produced it and the conversation that led to the code. This makes results reproducible by design, not by convention. There is no "I ran this in my notebook a month ago and I think these settings were right" — the entire chain of custody is preserved in the project artifact log.
Claude Science handles the logistics of where code runs. Local execution works for data exploration and lightweight modeling. For jobs that need more memory or GPU access, the workbench can dispatch tasks to a connected Linux box or HPC login node without you needing to manually SSH and set up environments. Modal integration brings GPU capacity on demand — useful for Boltz-2 structure predictions or Evo 2 sequence generation — and Modal is providing up to $2,000 in compute credits for selected AI for Science grant recipients.
The workbench ships with over 60 curated skills and connectors organized by discipline: genomics, single-cell, proteomics, structural biology, and cheminformatics. These are not generic API wrappers — they are purpose-built integrations that understand the data formats, common workflows, and domain conventions for each field. The NVIDIA BioNeMo Agent Toolkit brings Evo 2 (DNA foundation model), Boltz-2 (biomolecular interaction prediction), and OpenFold3 (protein structure prediction) directly into the workbench environment.
After you have generated results, the Reviewer Agent runs an independent pass over your work. It checks that citations actually support the claims made, validates numerical calculations, and flags potential errors before you share or submit. For labs producing long-form documents — manuscripts, grant sections, systematic reviews — the Reviewer Agent acts as a first-pass co-author check that catches mistakes that are easy to miss under deadline pressure.
Manifold Bio used Claude Science to nominate targets for tissue-targeting medicine experiments. The workbench assessed surface expression, trafficking behavior, and safety profiles for a panel of candidate proteins, then ranked them for experimental follow-up. Work that would have required a literature analyst, a bioinformatician, and a safety reviewer operating sequentially was compressed into a single Claude Science session.
Jérôme Lecoq, a neuroscientist at the Allen Institute for Brain Science, built a multi-agent "computational review template" inside Claude Science using approximately 20 custom skills. The system reads thousands of papers, extracts central claims, identifies contradictions across the literature, and builds narrative arcs suitable for a long-form scientific review. Reviews that previously took two years to complete are now produced in a fraction of that time. Lecoq has already produced around 10 reviews through the system, many exceeding 100 pages, with the Reviewer Agent checking every citation.
Stephen Francis, an epidemiologist at the UCSF Brain Tumor Center, used Claude Science to investigate the molecular epidemiology of glioma, specifically the genetic basis for germline variant susceptibility. The analysis that previously required a full research timeline was accelerated to approximately one-tenth of the previous time, and the results were independently validated. For clinical research programs where funding cycles and patient timelines impose hard constraints, that kind of acceleration is directly meaningful.
Claude Science is available at claude.com/science. Access requires an active Claude subscription at the Pro, Max, Team, or Enterprise tier. The workbench application runs on macOS and Linux.
For Team and Enterprise accounts, an admin must enable Claude Science before individual users can access it. Academic and nonprofit labs can apply for discounted Team plan seats through the AI for Science program.
Anthropic is supporting up to 50 research projects through its AI for Science initiative:
Applications are open to academic labs, nonprofit research institutes, and independent researchers working on scientific questions where Claude Science could provide meaningful acceleration.
Claude Science is best suited for researchers who regularly work across multiple tools and find the switching cost between them — literature search, coding, statistical modeling, documentation — to be the primary bottleneck in their workflow. Specific high-value cases include:
For teams that have already adopted Claude for general research tasks, Claude Science provides the structured environment and domain-specific tooling that turns ad hoc AI assistance into a repeatable research process. If you are interested in how Claude's broader agentic capabilities apply to other domains, see the top 10 use cases for Claude Fable 5.
Claude Science launched on June 30, 2026 and is currently in beta for Claude Pro, Max, Team, and Enterprise users. Academic and nonprofit labs can apply for discounted access and AI for Science project credits through July 15, 2026.
