Boston Comp-Bio Hack: Claude Science, Modal, Flagship

CAMBRIDGE, MA — On September 22, 2026, from 09:00 to 19:30 EDT, Anthropic, Modal, and Flagship Pioneering host a one-day computational biology hackathon. The challenge is specific: build cloud tools with Modal × Claude Science for autoresearch, molecular design, data scraping and processing, or any agent-automated heavy workload. That is a room and a compute stack. It is not a three-way product integration.

A secondary listing mirrors the date, the 09:00–19:30 EDT window, Cambridge, and free general admission. Luma still requires host approval to register and hides the street address until you do. Ultrathink is treating the event as the public intersection of three sourced facts: Anthropic’s science workbench, Modal’s user-billed burst GPUs, and Flagship’s Cambridge bioplatform roster. Co-hosting a hackathon is not the same as wiring the products together.

What the hosts put on the page

Luma’s copy is the primary. Spend a day building with the three teams in the room. Form a group or bring one. Demo at the end. Reception after. Audience: Greater Boston computational-biology engineers and academics. Credits: Modal and Claude. No dollar amounts on the event page.

The listed inspiration tracks, with the inputs and outputs the hosts wrote down:

• Protein function — FASTA in; predicted function, domains, homologs, structural insights, and confidence scores out. Large protein language models, structure prediction, homology and domain search.
• Drug discovery — disease, phenotype, or gene set in; ranked targets or drug candidates with scores and evidence out. Multimodal datasets, virtual screening, inference across genes and compounds.
• Imaging and phenotyping — microscopy in; phenotypes, embeddings, segmentation, and QC out. High-content datasets at cell or image scale.
• Biology knowledge graph — a gene, protein, disease, drug, or question in; predicted relationships, ranked hypotheses, and supporting paths out.

The brief above those tracks is the product constraint: start from a named life-sciences user and use Claude Science to build the missing tool. Modal is the cloud the job lands on when the laptop is not enough.

Claude Science is a workbench, not a model

Anthropic launched Claude Science on June 30, 2026. The company is explicit: a public-beta app, not a new model. It sits on the Claude the subscriber already pays for, on Pro, Max, Team, and Enterprise, on macOS and Linux. Day one: 60+ curated skills and connectors for genomics, single-cell, proteomics, structural biology, and cheminformatics. NVIDIA BioNeMo Agent Toolkit skills are in the box. The named models are Evo 2, Boltz-2, and OpenFold3.

Default execution is local. When a job needs a GPU or more CPUs than the machine has, the app proposes a Modal job. That billing split, and the protein-binder campaign written for the same harness, is prior Ultrathink coverage. This piece does not re-litigate the 1,440-binder release.

Modal bills the GPUs. Anthropic does not.

Modal’s June 30 post matches the Anthropic launch: the researcher connects their own Modal workspace; GPU and concurrent-CPU jobs route to Modal sandboxes; the researcher stays in Claude. Claude Science compute docs put the money in one paragraph: “You connect your account, jobs run on it, and Modal bills you directly. Anthropic doesn’t provide or bill compute and never sees a payment method.”

Each job shows a confirmation card with machine spec, per-second billing, and a maximum billable time. There is no spend ceiling inside Claude Science. Closing the app does not cancel a running Modal job. Modal’s launch post still says a new account includes $30 of free compute at signup. That figure is Modal’s, dated June 30, 2026.

A different pot, also from the June 30 Anthropic post: up to 50 AI for Science projects, up to $30,000 in Claude credits, plus up to $2,000 of Modal compute for select projects. Applications closed July 15, 2026. Awards were due by July 31. Project window: September 1 to December 1, 2026. Those credits are a closed launch program, not a standing offer, and they are not the unspecified “Modal and Claude credits” on the Luma page. Do not treat them as hackathon prize money.

Flagship is the Cambridge bioplatform host

Flagship Pioneering’s name on the Luma page is the sourced link to the hack. That is the claim. Flagship’s own “About” language, restated in the March 10, 2025 Lila unveil, is the thesis: it invents and builds bioplatform companies aimed at human health and sustainability. The same blurb: more than 100 scientific ventures since 2000, more than $60 billion in aggregate value, $14 billion of assets under management as of a July 2024 capital raise. The current roster listed there includes Moderna, Generate Biomedicines, Tessera, Inari, and Lila Sciences.

Lila is the Flagship company to name if you are writing about scientific superintelligence and autonomous labs. Flagship unveiled it on March 10, 2025. Founded in Flagship’s labs in 2023. $200 million in committed seed capital from Flagship plus General Catalyst, March Capital, ARK Venture Fund, Altitude Life Science Ventures, Blue Horizon Advisors, the State of Michigan Retirement System, Modi Ventures, and an ADIA subsidiary. Geoffrey von Maltzahn is CEO. Noubar Afeyan is chairman. Teams in Cambridge, San Francisco, and London. lila.ai describes an AI model plus “AI Science Factory” instruments that generate hypotheses, run experiments, and learn from the results.

“With Lila, our hypothesis is that by scaling experimentation, we can unlock emergent abilities and enable discoveries that remain hidden at smaller scales.”

Noubar Afeyan, Flagship Pioneering / Lila Sciences — March 10, 2025 press release

Lila’s own June 22, 2026 post on NVIDIA’s BioNeMo Agent Toolkit is a Lila–NVIDIA document. Ben Kompa writes that Lila trains agents with open BioNeMo models, proprietary Lila models, and NVIDIA ALCHEMI simulations, then grades hypotheses in physical labs. Claude Science also uses BioNeMo skills. Sharing a NVIDIA toolkit is not a Claude–Lila integration. No Flagship, Lila, Anthropic, or Modal primary in this file says the workbench talks to Lila’s factories.

Same city. Two different stacks.

The September 22 hack is the public place those three names share a calendar. The product stack on the brief is Claude Science plus a Modal workspace the builder connects and pays. Flagship is in the room as a Cambridge bioplatform host. Lila is Flagship context: a 2023 company with a 2025 seed and a closed-loop lab thesis. Keep those sentences apart. The hack does not announce a Lila connector. The Lila unveil does not mention Claude Science. Modal’s compute post does not mention Flagship.

If you are going, the sourced facts are the ones you can spend: a free, approval-gated Tuesday in Cambridge; four tracks with named inputs and outputs; Modal × Claude Science as the build constraint; credits of unnamed size; a demo and a reception. Everything else is someone else’s press release in the same metro area.