Anthropic is giving 10,000 scientists around the world free or heavily discounted access to Claude, the company announced on August 27, 2026 — the latest and broadest step yet in a year-long push to turn its AI models into infrastructure for scientific research.
The announcement bundles three things: a new Claude Team plan for scientists offering free and discounted subscription seats to verified researchers; an expansion of Anthropic’s existing AI for Science credit program beyond biology into every scientific field; and confirmation that a government-linked access program for Anthropic’s most restricted model tier, Mythos, has enrolled its first participants in life sciences research.
Taken together, the moves mark a shift from general-purpose AI marketing toward something more specific: positioning Claude as scientific infrastructure that labs build workflows around, not just a chatbot researchers happen to use.
Table of Contents
What Anthropic Announced
The centerpiece is the Claude Team plan for scientists. Verified principal investigators — or an equivalent role — at academic or nonprofit research institutions can now sign up for a Claude Team subscription and add researchers from their lab. Standard seats are free; Premium seats, which come with five times the usage limits, cost $15 per month. Both tiers are promotional pricing locked in for 12 months from either August 27, 2026, or the date a group signs up.
Anthropic says it intends to expand the program “well beyond” the initial 10,000 seats over the coming months, though that is a stated intention rather than a completed step.
Alongside the subscription plan, Anthropic is broadening its AI for Science program, which has offered free API credits to researchers since May 2025. Until now, that program was described as “largely” focused on biological sciences. It’s now open to researchers in any scientific field, with awards of up to $50,000 in credits per project for labs that outgrow what a standard or premium subscription seat provides.
The third and most sensitive element is a life sciences access program for Mythos-class models — Anthropic’s most capable and most tightly restricted model tier. The company says it is “working in partnership with the US government” to give life sciences professionals access to these models for research and development, and that it has “enrolled our first participants.” Anthropic did not name the government agency involved, how many participants have joined, or when broader access might follow.
It’s worth being precise about what this program is not. It is not the same as the export-control episode that suspended access to Anthropic’s Fable and Mythos models entirely in June 2026 — that involved the US Department of Commerce and was about cybersecurity and national security, not life sciences. The agency behind this new biology-focused access program has not been disclosed.
Why It Matters
None of this happens in isolation. OpenAI launched a comparable “ChatGPT for Academic Researchers” program earlier this summer, initially offering 10,000 seats and planning to scale to 100,000 by 2027, selecting some participants through a lottery among more than 13,000 first-wave applicants. Google DeepMind, through its Isomorphic Labs subsidiary and a “Gemini for Science” platform built around AlphaFold, is pursuing a different strategy centered on proprietary molecular-design models rather than broad subscription access.
Anthropic’s announcement lands squarely in that same window, which makes it as much a competitive move as a philanthropic one. Frontier AI labs are converging on the same idea at nearly the same time: getting research institutions hooked into a company’s tools early, cheaply, is a way of shaping how an entire generation of scientific work gets done.
This announcement is also the latest step in a buildup Anthropic has been assembling for more than a year:
| Date | Development |
|---|---|
| May 2025 | AI for Science program launches, with a biology and life-sciences focus |
| April 2026 | Project Glasswing launches; Mythos Preview released to a limited group of cyberdefenders |
| June 9, 2026 | Fable 5 and Mythos 5 launch |
| June 12–July 1, 2026 | Fable/Mythos access suspended over national-security concerns, then partially restored under Commerce Department authorization |
| June 30, 2026 | Claude Science launches in beta |
| July–August 2026 | Claude Science project grants and a rare-disease credit track run their course |
| August 27, 2026 | Claude Team plan for scientists and AI for Science expansion announced |
How the Claude Team Plan for Scientists Works
The plan is structured around verification. A principal investigator applies through a verification form, confirms their institutional affiliation, and once approved can invite lab members onto Standard or Premium seats. Anthropic caps groups at 1 to 25 seats by default, with larger labs able to contact the company directly.
The pricing is explicitly promotional, not permanent: after 12 months, seats renew at whatever the standard price for the plan is at that time, and Anthropic says it will notify administrators before that happens. Groups that go 90 days without activity may also be returned to standard Team pricing, freeing the discounted seats for active researchers.
The AI for Science Credit Program Expansion
The credit program is a separate track from the subscription plan, aimed at labs whose computational needs exceed what a seat provides. It predates this announcement by more than a year — Anthropic launched it in May 2025 with a stated focus on biology and life sciences, citing goals like accelerating drug discovery and understanding complex biological systems.
Since then, the program has run in phases. A Claude Science-specific round in mid-2026 funded up to 50 projects with up to $30,000 in credits each. A separate rare-disease-focused track ran from late July through early August 2026, offering up to $50,000 in credits over six months to teams working on genetic disease mechanisms and early-stage treatment development.
This announcement expands the program’s scope beyond biology to cover any scientific field, still offering up to $50,000 in credits per project. Anthropic has not yet named which researchers or projects have been awarded credits under this newly widened scope — the expansion is too recent for that.
Claude Science — The Underlying Workbench
Much of what these scientists will actually use is Claude Science, an AI research workbench Anthropic launched in beta on June 30, 2026. It’s worth being clear about what this is: not a new model, but an application layered on top of Claude’s existing models, built to bring together the fragmented tools researchers normally juggle — literature databases, coding environments, HPC clusters, and specialized scientific software.
Claude Science runs locally on macOS or Linux, or remotely over SSH on a lab’s own infrastructure, and comes pre-configured with more than 60 skills and connectors for genomics, single-cell analysis, structural biology, and cheminformatics. It integrates with NVIDIA’s BioNeMo toolkit to reach specialized models like Evo 2, Boltz-2, and OpenFold3, and includes a “reviewer agent” designed to check citations and calculations for errors.
Early users cited by Anthropic include Manifold Bio, which used the tool to help nominate drug targets; the Allen Institute, where a researcher built a multi-agent pipeline that reportedly cut long-form literature reviews from as long as two years down to a fraction of that time; and UCSF’s Brain Tumor Center, where a lab studying the genetic basis of glioma said the tool cut germline variant analysis time to roughly one-tenth of what it previously took, with results the lab independently validated itself. These are early pilot deployments, not evidence of large-scale commercial adoption.
Model Tiers — Opus, Fable, and Mythos Explained
Understanding what scientists can and can’t do requires understanding Anthropic’s model hierarchy, which the company laid out when it launched Fable 5 and Mythos 5 on June 9, 2026.
Opus-class models are Anthropic’s standard, generally available top tier — what most subscribers, including scientists on the new Team plan, currently use for biology and chemistry work. Above that sits a newer category Anthropic calls “Mythos-class,” which includes two models: Fable 5 and Mythos 5.
Here’s the part that trips people up: Fable 5 and Mythos 5 are the same underlying model. The difference is which safeguards are switched on.
| Model | Availability | Safeguards | Pricing |
|---|---|---|---|
| Opus-class (e.g., Opus 4.8, Opus 5) | Generally available | Standard safeguards | Standard subscription/API pricing |
| Fable 5 | Generally available | Automatic fallback to Opus 4.8 on sensitive cybersecurity, biology/chemistry, and distillation-related queries (affects under 5% of sessions) | $10 per million input tokens / $50 per million output tokens |
| Mythos 5 | Restricted to vetted users (Project Glasswing partners; a small group of life sciences researchers) | Same underlying model as Fable 5, with select safeguards lifted for approved use cases | $10 per million input tokens / $50 per million output tokens |
Fable 5, released for general use, comes with safety classifiers that automatically reroute certain queries — related to cybersecurity, biology and chemistry, or attempts to extract the model’s capabilities — to the less capable but still strong Opus 4.8 instead of answering with Fable’s full capability. Mythos 5 is the same model with some of those safeguards lifted, but it’s only available to vetted users: cybersecurity partners in Anthropic’s Project Glasswing program, and — as of this announcement — a small, unnamed group of life sciences researchers.
The Mythos Life Sciences Access Program
This is the part of the announcement that deserves the most caution. Anthropic says it is working with the US government to give life sciences professionals access to Mythos-class capabilities, with biology and chemistry safeguards lifted but cybersecurity safeguards still in place. The company says it has enrolled its first participants and expects to “share more and increase access soon.”
What’s confirmed: the program exists, and it has started, in some limited form. What isn’t confirmed: which government agency is involved, how many participants have joined, or any timeline for wider access. This is a separate arrangement from the widely reported June 2026 episode in which the Commerce Department directed Anthropic to suspend all Fable and Mythos access over national-security concerns, later authorizing a partial restoration for vetted US organizations working on cybersecurity and critical infrastructure. That earlier episode involved named officials and a documented paper trail; this new program, so far, does not.
Why Biology and Chemistry Remain Restricted
Anthropic has been explicit that the restriction isn’t a technical limitation — it’s a deliberate safety choice. The company argues that Mythos-class capabilities in biology pose a genuine dual-use risk: the same reasoning that helps a legitimate researcher design a therapeutic protein could, in principle, help someone with bad intentions design something harmful.
Anthropic has pointed to its own testing as evidence of both the promise and the risk. In one evaluation, Mythos-class models were tested on predicting how genetic modifications would affect the assembly of a viral shell in adeno-associated viruses — a technique relevant to legitimate gene-therapy research, and one where the models reportedly outperformed specialized protein-language tools despite not being trained specifically for the task. Anthropic has treated that result as evidence for keeping biology capabilities gated rather than an argument for opening them up.
Evidence of Claude’s Scientific Capability
That safety caution sits alongside notable results Anthropic published this month. In a wet-lab-validated protein design campaign using Mythos Preview and Opus 4.8, Claude succeeded on 14 of 15 targets, with hit rates of roughly 23–35% against a typical industry rate of 10–15%. Two outside organizations, Adaptyv Bio and Twist Bioscience, independently produced and tested the designs in their labs.
In a separate test, Claude Opus 5 was given raw NMR and mass-spectrometry files from a contract lab and, in a two-sentence prompt, asked to determine a compound’s identity and purity. It returned results in 19 to 23 minutes that closely matched the lab’s own findings, including a purity measurement of 96.4% against the lab’s 96.33%.
Anthropic also reported that an unreleased research version of Claude improved a long-standing mathematical bound related to the Riemann hypothesis, raising the known proportion of zeta-function zeros satisfying the hypothesis from 41.6% to 67.2%, reviewed by two outside mathematicians and formalized in a verified proof.
These are real, specific results — but they are company-run experiments with limited outside review, not standard peer-reviewed publications. Read them as promising early evidence, not settled consensus.
How This Compares to Competitors
Anthropic isn’t alone in courting researchers, and it isn’t first. OpenAI’s ChatGPT for Academic Researchers program, launched over the summer, offers free model access at a similar initial scale and has already drawn more applications than it can immediately accommodate. Google DeepMind, through Isomorphic Labs and its Gemini for Science platform, is pursuing a different bet built around AlphaFold and other purpose-built molecular models rather than a subscription discount.
| Anthropic | OpenAI | Google DeepMind | |
|---|---|---|---|
| Program | Claude Team plan for scientists | ChatGPT for Academic Researchers | Isomorphic Labs / Gemini for Science |
| Initial scale | 10,000 seats | 10,000 seats, scaling toward 100,000 by 2027 | Not a comparable subscription program |
| Pricing | Free Standard / $15 per month Premium, 12-month promotional pricing | Not publicly disclosed in comparable terms | Not applicable — model/platform access, not a subscription discount |
| Selection | PI verification | Lottery among first-wave applicants (13,000+) | N/A |
The three approaches aren’t quite the same product. Independent comparisons describe Claude Science as an orchestration layer stitching together databases, compute, and specialist tools, rather than a domain-specific model competing directly with something like AlphaFold. Literature-focused tools like Elicit and Consensus occupy a narrower niche — structured evidence synthesis and citation search — that researchers may use alongside, rather than instead of, a broader workbench like Claude Science.
Risks, Limitations, and Open Questions
A few limitations are worth keeping in view. Fable 5’s safety classifiers, by Anthropic’s own account, are tuned conservatively enough to occasionally block harmless requests, and the company has said it’s working to reduce those false positives over time. The June 2026 suspension of Fable and Mythos access — triggered by government national-security concerns and only partially reversed — is a reminder that access to these restricted models has already proven unstable once, even if the current life-sciences program is a distinct arrangement.
More broadly, no AI-discovered drug, from Anthropic or any competitor, has yet completed a large-scale clinical trial. Protein-design and analytical-chemistry results, however striking, sit far upstream of that point in the drug-development pipeline.
What’s Still Unknown
Several details remain undisclosed. Anthropic hasn’t named the government agency involved in the Mythos life-sciences program, how many researchers have been enrolled, or when broader access might open up. The company also hasn’t identified which specific projects or institutions will receive credits under the newly expanded, all-fields version of the AI for Science program. And because the subscription plan is barely a day old at the time of this writing, there’s no usage or adoption data yet to assess how it’s actually being taken up.
Conclusion
What’s solid here is the subscription plan itself: a concrete, verifiable offer of free and discounted Claude access to a defined group of scientists, backed by specific pricing and eligibility terms. What’s much less clear is the life sciences Mythos program — real, but still small, unnamed, and unproven at scale. Readers should watch for two things in the coming months: whether Anthropic actually expands the 10,000-seat program as promised, and whether it discloses more about who its government partner is and how the trusted-access model for biology actually works in practice. Until then, the responsible read is that Anthropic has opened a wide, low-friction front door for most scientific work, while keeping a narrow, closely monitored side door for the riskiest capabilities.
Frequently Asked Questions
Who is eligible for Anthropic’s Claude Team plan for scientists? Principal investigators, or an equivalent role, at academic or nonprofit research institutions can apply through a verification form. Once verified, they can add researchers from their lab, typically up to 25 seats per group, with larger labs able to contact Anthropic directly.
How much does the Claude Team plan for scientists cost? Standard seats are free. Premium seats, which include five times the usage limits, cost $15 per month. Both are promotional prices locked in for 12 months from either August 27, 2026, or a group’s sign-up date, after which they renew at standard pricing.
What is Anthropic’s AI for Science program, and how do researchers apply? It’s a separate credit program, running since May 2025, that offers qualifying research projects up to $50,000 in Claude credits. Originally focused on biology and life sciences, it has now expanded to cover all scientific fields. Applications are typically submitted through a form and reviewed by Anthropic staff, including subject-matter experts.
What is the difference between Claude Opus, Fable 5, and Mythos 5? Opus-class models are Anthropic’s standard generally available tier. Fable 5 and Mythos 5 are the same underlying “Mythos-class” model, but Fable 5 runs with safety classifiers that reroute sensitive cybersecurity, biology, and chemistry queries to Opus 4.8, while Mythos 5 has some of those safeguards lifted for a small group of vetted users.
Can biologists and chemists use Claude Mythos 5? Only through a limited, still-early trusted-access program that Anthropic says it’s running in partnership with the US government, which has enrolled its first participants. The general public and most researchers on the new Team plan remain limited to Opus-class models for biology and chemistry work.
Is Anthropic’s scientist program similar to OpenAI’s academic researcher program? Yes, in scale and timing. OpenAI’s ChatGPT for Academic Researchers program launched earlier in the summer of 2026 with an initial 10,000 seats and plans to reach 100,000 researchers by 2027. Google DeepMind is pursuing a different strategy centered on proprietary scientific models like AlphaFold rather than a comparable subscription discount.
What risks does Anthropic cite for restricting biology and chemistry queries? Anthropic says advanced biological reasoning capabilities carry dual-use risk — the same capabilities that could accelerate legitimate drug design could, in principle, provide meaningful assistance to someone attempting to cause harm. The company has cited its own testing, including work on viral shell assembly prediction, as evidence for keeping these capabilities restricted rather than generally available.