Anthropic launched Claude Science on June 30, 2026, an AI workbench enabling life sciences researchers to run computational workloads directly from conversations with Claude. This platform integrates with Modal, providing on-demand, scalable compute resources for data processing, protein modeling, and molecule design, according to Gizmodo.
This integration addresses the significant compute demands of modern biological research. Many tasks in life sciences, from running inference on protein language models to screening millions of compounds, are highly resource-intensive and require infrastructure that can scale dynamically.
By linking Claude Science with Modal, researchers gain access to a powerful backend for bursty, heterogeneous workloads. This approach helps scientific discovery by making advanced AI and computational tools more accessible and efficient.
Computational Biology's Compute Challenge
Traditional computational biology workflows are often fragmented, relying on a mix of local machines, HPC clusters, and cloud virtual machines, leading to inefficiencies in resource provisioning and environment setup. Researchers frequently navigate complex processes like SSHing into servers and managing diverse bioinformatics tools, which slows down the scientific iteration cycle for compute-heavy tasks.Claude Science simplifies this by allowing code execution within the conversational interface. While early-stage analysis can occur in a sandboxed environment on a researcher's machine, demanding computational biology workflows quickly outgrow these initial setups.
Modal's Role in Scaling Scientific AI
Modal enhances Claude Science by providing elastic, reproducible infrastructure that automatically scales compute resources to match workload demands, supporting computationally intensive tasks without manual configuration. It enables researchers to connect their Modal workspace, automatically routing GPU or multi-CPU workloads to Modal sandboxes, ensuring optimal hardware utilization for each step of a biology pipeline.The platform facilitates parallel processing, or "fan out," allowing tasks like virtual screening against large compound libraries to complete in minutes rather than hours. This capability is crucial for accelerating research that involves extensive data analysis.
Modal also offers fine-grained GPU access, so hardware is only requested for specific functions within a pipeline, not for the entire workflow. Shared storage via Modal Volumes ensures datasets and model checkpoints are accessible across all jobs, eliminating expensive data movement. Additionally, Modal Images support reproducible environments, making dependency management consistent and reliable across runs.
We have a lot of different models that give us different layers of insights on these proteins, and being able to run them all on the hardware that makes sense is what makes the product possible.Kevin Wu, a Machine Learning Researcher, highlights Modal's ability to efficiently handle diverse models on appropriate hardware. He noted the platform's speed and ease for bursty workloads, stating, "Sometimes we spin up hundreds of GPUs at a time, and the fact it's up in a few minutes without onerous configurations or dashboards whenever we need to is kind of a miracle."
— Kevin Wu, Machine Learning Researcher
| Feature | Traditional Workflow | Claude Science with Modal |
|---|---|---|
| Compute Provisioning | Manual setup, HPC queues | Automatic, on-demand scaling |
| GPU Access | Fixed cluster allocation | Per-step, dynamic allocation |
| Data Management | Expensive data movement | Shared storage (Modal Volumes) |
| Environment Setup | Inconsistent, dependency hell | Reproducible (Modal Images) |
| Parallelism | Sequential or complex config | Automated fan out across containers |
Impact on Research Workflows
The integration empowers researchers to conduct complex experiments directly through natural language prompts, simplifying tasks from predicting protein structures to designing genome-wide screens. This streamlines workflows, allowing scientists to focus more on scientific inquiry and less on infrastructure management, leading to faster hypothesis testing and discovery cycles.For example, a researcher can prompt Claude to predict a protein's 3D structure using a Modal GPU, receiving the output and confidence scores within the conversation. Engineering enzymes across multiple models, such as using ESMFold, ESM-2, and ProteinMPNN, also becomes feasible, with Modal providing the varied hardware needed for each model family.
Highly parallel tasks like designing genome-wide CRISPR knockout screens, which involve checking thousands of guide candidates against the human genome, can be fanned out across hundreds of Modal containers and completed in minutes. Similarly, complex single-cell analysis at dataset scale becomes manageable, enabling rapid identification of immune population changes in studies. This robust compute infrastructure represents a significant advancement for AI-assisted development in biology, as previously discussed in Claude Boosts Limits, Inks SpaceX Compute Deal.
To support academic research, Anthropic is committing up to $100,000 in compute credits for its AI for Science Claude Science Cohort. Projects can receive $500–$2,000, with applications open through July 15, 2026, and awards announced by July 31. Projects will run from September 1 to December 1, 2026.
