# ProGen3 & OpenCRISPR by Profluent

Canonical page: https://technooptimist.io/ai/progen3
Frontier: [AI](https://technooptimist.io/ai.md) · Field: Biology models · Company: [Profluent](https://technooptimist.io/companies/profluent.md)
Stage: Pilot (3 of 5: Research → Proto → Pilot → Deploy → Scaling)
Updated: last update 2026-04-28, facts checked 2026-10-11
Goal: Write new proteins to order instead of borrowing them from nature, starting with gene editors that are smaller, more precise or able to insert whole genes, so more genetic diseases become fixable.
Status: Partnered with Lilly (up to $2.25B) and BioMarin; designs still preclinical.
Next milestone: First AI-designed therapy dosed in humans

## Milestones

| Status | Milestone | Date | Slip |
| --- | --- | --- | --- |
| Done | OpenCRISPR-1 released openly | Apr 2024 |   |
| Done | ProGen3 shows protein scaling laws | 16 Apr 2025 |   |
| Done | OpenCRISPR-1 published in Nature | 30 Jul 2025 |   |
| Done | Eli Lilly recombinase deal | 28 Apr 2026 |   |
| Done | BioMarin partnership | 30 Sep 2026 |   |
| Planned | First AI-designed therapy dosed in humans |   |   |

## Most important updates

- **28 Apr 2026**: Eli Lilly signs deal worth up to $2.25B with Profluent for AI-designed gene editors. Lilly partnered with Profluent to design recombinases, enzymes that could insert entire genes into a patient's DNA rather than snip or swap letters. ([our story](https://technooptimist.io/stories/eli-lilly-signs-deal-worth-up-to-2-25b-with-profluent-for-ai-designed-f96fe9f4.md) · [original source](https://www.statnews.com/2026/04/28/eli-lilly-crispr-gene-editing-deal-profluent-ai/))
- **16 Apr 2025**: Profluent unveils ProGen3, reporting scaling laws for protein-design models. Profluent said its ProGen3 models follow scaling laws: larger models trained on more protein data generate more diverse proteins that still work in the lab and adapt faster to lab feedback. ([our story](https://technooptimist.io/stories/profluent-unveils-progen3-reporting-scaling-laws-for-protein-design-cedc929e.md) · [original source](https://fortune.com/2025/04/16/biotech-profluent-ai-scaling-laws-protein-design-models-opencrispr-openantibodies/))

## Current obstacles

- **Delivery into the body**: Even a better editor must be packaged into a virus or lipid particle and reach the right cells; smaller editors help.
- **Lab testing is the bottleneck**: Models write proteins in seconds, but each design still needs weeks of wet-lab work to confirm it works.

## Physics limits

- **Sequence is not function**: A model trained on sequences infers function indirectly; whether a new protein works in a cell can only be proven by experiment.
- **The space is astronomically large**: A 300-amino-acid protein has 20^300 possible sequences; models sample a sliver close to what evolution already explored.

## How it works

### Read: Learn the language of proteins

The model reads billions of natural protein sequences and learns which strings of amino acids fold and work.

### Write: Generate new sequences

Asked for a family like CRISPR editors, it writes new members hundreds of mutations from anything in nature.

### Test: Lab feedback

Designs are built and tested; results steer the model toward traits like stability, binding or editing precision.

### License: Partners take it to medicine

Profluent does not run its own drug pipeline; partners like Lilly develop the editors into treatments.

## Spec sheet

| Spec | ProGen3 & OpenCRISPR | Provenance |
| --- | --- | --- |
| Largest model trained on | About 3.4 billion protein sequences | reported |
| Protein Atlas size | 115+ billion unique proteins | reported |
| OpenCRISPR-1 distance from SpCas9 | About 400 mutations | reported |
| CRISPR operons mined for training | 1 million+ | reported |

Provenance: reported = stated by the company; estimated = our estimate; sample = a sample figure.

## Papers and demos

- **30 Jul 2025** (paper): [Design of highly functional genome editors by modelling CRISPR–Cas sequences (Nature)](https://www.nature.com/articles/s41586-025-09298-z). Peer-reviewed proof that an AI-written editor, OpenCRISPR-1, works in human cells.
- **Apr 2025** (paper): [Scaling unlocks broader generation and deeper functional understanding of proteins (ProGen3)](https://www.biorxiv.org/content/10.1101/2025.04.15.649055v2). Evidence that protein models follow scaling laws.

## About Profluent

Profluent trains large language models on protein sequences to design new proteins, starting with gene editors. Its OpenCRISPR-1 was the first AI-designed CRISPR editor to edit human DNA.

- ProGen3 showed protein models get better with scale, like chatbots: bigger models make more varied working proteins.
- Eli Lilly deal (April 2026) worth up to $2.25B in milestones to design recombinases that insert whole genes.
- Raised $106M Series B (Altimeter, Bezos Expeditions) in late 2025; about $150M in total.

- Headquarters: US
- Status: private
- Raised: $106M
- Last round: Series B $106M, Nov 2025, led by Altimeter Capital, Bezos Expeditions
- Website: https://www.profluent.bio
- People: Ali Madani (Founder & CEO)
- Partners, customers, investors: Eli Lilly (partner); BioMarin (partner); Corteva (partner); Integrated DNA Technologies (partner)

Company page: https://technooptimist.io/companies/profluent.md

## Related programs

Everything in AI: https://technooptimist.io/ai.md

---

The Techno Optimist tracks every frontier of technology, program by program: https://technooptimist.io/ (a Markdown version of any page: add .md to its URL; site map for agents: https://technooptimist.io/llms.txt).
Data API (JSON, realtime stream): https://technooptimist.io/developers. People read 5 pages free, then Reader $10/month; these Markdown pages are open to agents.