Chai Discovery Review: Features, Pricing, Pros, and Cons

If you work anywhere near drug discovery or computational biology, you've likely seen the name pop up in the last year, which is exactly why this Chai Discovery review exists — to cut through the funding headlines and Eli Lilly press releases and explain what the platform actually does, who can access it, and whether it's worth pursuing for your lab or company. Unlike most AI tools that anyone can sign up for with a credit card, Chai Discovery operates in a different world entirely: biologics R&D, enterprise pharma partnerships, and access that isn't self-serve. That changes how you should evaluate it.

Short answer: Chai Discovery is one of the most credible AI drug-design companies operating today, with results that have convinced Eli Lilly and a long list of top-tier investors to bet on it. But its flagship model isn't something you can just try out — access is selective, pricing isn't public, and it's built for organizations doing serious biologics work, not hobbyists or students. Here's the full picture.

What Is Chai Discovery?

Chai Discovery is a San Francisco-based AI company founded in 2024 that builds frontier models to predict and design the interactions between biochemical molecules — proteins, antibodies, small molecules, DNA, and RNA. Its stated mission is to turn biology from a purely experimental science into an engineering discipline, where designing a working therapeutic molecule becomes closer to a computational problem than a years-long trial-and-error lab process.

The company was founded by Joshua Meier (CEO), Jack Dent (President), Matthew McPartlon, and Jacques Boitreaud. The founding team's backgrounds read like a who's-who of frontier AI and biotech: Meier previously worked at OpenAI and Meta's generative biology group before becoming Chief AI Officer at antibody developer Absci, while Dent spent years at Stripe building product and engineering teams. The broader team includes alumni from OpenAI, Meta FAIR, Google X, and Stripe.

That pedigree attracted serious capital fast. Chai raised a $70 million Series A in mid-2025 led by Menlo Ventures, followed by a $130 million Series B in December 2025 co-led by Oak HC/FT and General Catalyst, valuing the company at $1.3 billion. Total funding now sits north of $225 million, with backers including OpenAI itself, Thrive Capital, Dimension, Emerson Collective, Glade Brook Capital, and SV Angel. Despite that valuation, the company remains lean — around 28 to 29 employees as of early 2026.

The moment that turned Chai from "promising AI biotech startup" into a headline name was its January 2026 collaboration with Eli Lilly, one of the pharma industry's largest AI software deals to date. Under the agreement, Lilly is deploying Chai's platform to design novel biologic therapeutics across multiple targets, and Chai is building a custom model trained exclusively on Lilly's proprietary data to fit Lilly's internal discovery workflows.

Core Products: Chai-1 and Chai-2

Chai Discovery's technology comes in two very different flavors — one open to anyone with the right hardware, and one that's tightly gated.

Chai-1: Open-Source Structure Prediction

Chai-1 is a multimodal foundation model for biomolecular structure prediction, released as open source on GitHub in 2024. It performs unified prediction across proteins, small molecules, DNA, RNA, glycosylation, and covalent modifications, and its benchmark performance is competitive with DeepMind's AlphaFold 3 on structure prediction tasks.

Practically, Chai-1 runs as a Python package (chai_lab) on Linux with Python 3.10 or later, and it needs a CUDA-capable GPU with bfloat16 support — the company recommends an A100 80GB, H100 80GB, or L40S 48GB card, though users have reported it running on consumer-grade RTX 4090 hardware for smaller complexes. There's no cost to use it, but there is a real compute barrier: this isn't something you casually run on a laptop.

Chai-2: Zero-Shot Antibody Design

Chai-2 is the platform's more dramatic leap — a generative model that designs full antibodies from scratch, using only a target epitope as input, with no need for the iterative lab screening rounds that traditionally define antibody discovery. Meier has described it as "Photoshop for proteins," and the benchmark numbers back up the ambition:

• A reported global success rate of roughly 20% for de novo antibody design across around 50 tested targets — compared to a typical success rate below 0.1% for conventional computational screening approaches, a roughly 100-fold improvement
• Structure prediction accuracy reaching DockQ scores above 0.8 for 34% of antibody-antigen complexes, roughly double Chai-1's earlier performance
• A published case where a target that had previously consumed more than $5 million in traditional R&D spend was solved in a matter of hours of compute time and validated in the lab within two weeks
• Support for multiple antibody formats, including nanobodies (VHHs) and full VH-VL structures, with extensions toward bispecifics, antibody-drug conjugates, and CAR-T constructs
• A reported 68% wet-lab success rate when the model was challenged to design miniprotein binders, with picomolar-level binding affinities in several cases

Those numbers are genuinely unusual for the field, and they're a big part of why Eli Lilly, and a growing list of biopharma and academic partners, have taken notice.

Platform Access and How It Actually Works

This is the part that trips people up if they come in expecting a typical SaaS product. Chai-2 is not publicly available. There's no public web server, no open API, and no downloadable version — access is limited to early-access partners selected through Chai Discovery's Responsible Deployment Framework, a policy that prioritizes researchers and organizations working on molecules intended to benefit human health.

In practice, the workflow looks like this for a partner organization:

• An interested academic lab, biotech, or pharma company applies for access through Chai directly
• Chai evaluates the request under its responsible-deployment criteria before granting access
• Approved partners can submit a target epitope and receive generated antibody candidates, moving from computational design to lab validation in as little as two weeks
• Larger partners, like Eli Lilly, can go further and commission a custom model trained specifically on their own proprietary data, tailored to their internal discovery pipeline

Chai-1, by contrast, is fully open — anyone can pull the code from GitHub and run it on their own infrastructure, which gives academic labs and smaller biotechs a real (if compute-intensive) way to engage with the company's technology without needing a partnership deal.

Chai Discovery Pricing

There is no published, self-serve pricing page for Chai Discovery's flagship platform, and that's a deliberate structural choice, not an oversight. Here's how the two products break down on cost:

For most organizations, that means budgeting for Chai Discovery isn't a matter of picking a subscription tier — it's a matter of reaching out, going through the responsible-deployment application process, and negotiating terms directly, likely alongside legal and data-sharing agreements given the sensitivity of biologics R&D. Expect this to look more like a strategic pharma partnership than a software purchase, both in timeline and in dollar figures.

What Chai Discovery Does Well (Pros)

• Chai-1 is genuinely free and state-of-the-art. Any lab with access to a capable GPU can run a structure prediction model that holds its own against AlphaFold 3, at zero licensing cost.
• Chai-2's hit rates are a real scientific leap, not just marketing. A reported 20% zero-shot success rate against a field where under 0.1% is typical is the kind of gap that changes how discovery timelines get planned.
• Speed and cost savings are demonstrated, not theoretical. The example of solving a $5 million R&D problem in hours of compute, validated in two weeks, is a concrete result rather than a projected benchmark.
• Elite technical pedigree and serious capital backing. A founding team from OpenAI, Meta FAIR, and Absci, combined with $225M+ raised from OpenAI, Thrive Capital, and General Catalyst, signals a level of technical and financial credibility that's rare even in a crowded AI-biotech field.
• Format flexibility beyond simple antibodies. Support for nanobodies, full VH-VL antibodies, and movement toward bispecifics, ADCs, and CAR-T constructs means the platform isn't a one-trick tool.
• Real pharma validation, not just academic benchmarks. The Eli Lilly collaboration, one of the largest AI software deals in pharma to date, is a strong signal that the technology performs under real commercial scrutiny, not just in controlled demos.
• Custom model training for large partners. Organizations with substantial proprietary data, like Lilly, can get a model tuned specifically to their own discovery workflows rather than a generic tool.

Where Chai Discovery Falls Short (Cons)

• Chai-2 is not accessible to most people who want it. Unless you're selected through the Responsible Deployment Framework, there's no way to try the flagship product — no free trial, no public API, no self-serve tier.
• Pricing opacity makes budgeting difficult. Without published rates, smaller biotechs and academic groups can't easily estimate cost before entering a lengthy application and negotiation process.
• The access model favors large, well-resourced organizations. Custom model deals like the Lilly partnership are realistically out of reach for smaller labs, even if their science is compelling, simply due to negotiating leverage and budget.
• Chai-1 has a real compute barrier. Free doesn't mean accessible — running it properly requires A100, H100, or similarly capable GPU hardware, which many academic and early-stage teams simply don't have on hand.
• Limited public disclosure on Chai-2's architecture. Full training and architectural details for Chai-2 haven't been made public, which makes independent, third-party evaluation of its claims harder than it would be for a fully open model.
• AI-generated designs are a starting point, not a finished drug. Even a validated hit rate of 20% still requires wet-lab confirmation, preclinical testing, and the full regulatory pathway — Chai-2 accelerates early discovery, it doesn't skip the rest of drug development.
• Young company, still scaling operational maturity. At under 30 employees managing partnerships with organizations the size of Eli Lilly, there's real execution risk in scaling support and infrastructure alongside demand.

Chai Discovery vs. the Alternatives

Chai Discovery sits in a fast-moving field of AI-driven molecular design companies, each taking a slightly different approach.

The honest comparison: if you need free, general-purpose structure prediction and have the compute to run it, Chai-1 or AlphaFold 3 are both strong, accessible choices. If you specifically need de novo antibody design and can get through Chai's access process, Chai-2's reported hit rates currently outpace what publicly available alternatives claim — the tradeoff is that "publicly available" isn't the right description for it yet.

Who Should Pursue Chai Discovery in 2026?

Chai Discovery is a strong fit for:

• Mid-size to large biopharma companies with active biologics discovery programs and the resources to negotiate an enterprise partnership
• Academic and industry research labs with access to serious GPU infrastructure who want a free, high-performing structure prediction tool via Chai-1
• Organizations specifically targeting antibody, nanobody, or emerging biologic formats like bispecifics and ADCs
• Companies with substantial proprietary discovery data who could benefit from a custom-trained model, similar to the Eli Lilly arrangement

It's a weaker fit for:

• Small academic labs or startups without GPU infrastructure or the leverage to secure early access to Chai-2
• Teams expecting transparent, self-serve pricing they can budget for without a sales conversation
• Anyone expecting a finished therapeutic rather than an accelerated starting point still requiring full preclinical and clinical development
• Researchers who need full architectural transparency for peer review or independent replication of Chai-2's specific claims

Final Verdict: Chai Discovery Review Conclusion

Wrapping up this Chai Discovery review — is the platform worth pursuing in 2026? For the right organization, yes, and the Eli Lilly partnership alone suggests the industry's most demanding buyers already agree. The reported zero-shot antibody design hit rates are a legitimate step change from what traditional and prior computational approaches have delivered, and Chai-1's open-source availability means even resource-constrained labs can engage with serious structure prediction technology at no licensing cost.

Where you need realistic expectations is access and cost. Chai-2 isn't a tool you sign up for on a Tuesday afternoon — it's a partnership you apply for, and pricing is a negotiation rather than a published number. If your organization has the biologics program, the data, and the scale to make that conversation worthwhile, Chai Discovery is one of the most credible bets in AI-driven drug design today. If you're a smaller team just looking to experiment, start with Chai-1 on your own infrastructure, watch how the company expands access over the next year, and revisit Chai-2 when — or if — its availability broadens beyond selective enterprise partnerships.