Artificial intelligence · Protein design
Anthrogen
A San Francisco lab that set out to engineer carbon-eating bacteria and has become a protein foundation model company, now shipping a 102 billion parameter model called Odyssey to early access users.
Two companies in one name
Search for Anthrogen today and two different companies come back. One, from late 2024, grows genetically modified photosynthetic bacteria that eat carbon dioxide and feeds their output through AI-designed enzymes to make jet fuel and plastics at, the company said, 80 percent of the cost of the fossil fuel version. The other, from late 2025 onward, trains a 102 billion parameter language model on protein sequences and structures and calls itself “an AI research lab for the biological sciences.” Both companies are the same three letters, same city, same CIK-free legal shell. The name has not changed since early 2024. What Anthrogen does under that name has changed twice.
The problem, twice redefined
The company started in 2023 as Arctic Capture, built around a single claim: it had engineered the fastest known photosynthesizing organism on the planet, a bacterium that could be pushed to consume carbon dioxide and sunlight fast enough to make chemical feedstocks commercially. By November 2024, when it announced a $4 million seed round, the pitch had widened. Enzymes designed by the company’s AI models would turn that captured carbon into fuels, plastics, starches and cellulose, undercutting petroleum-based versions on cost rather than asking customers to pay a “green premium.” CEO Ankit Singhal told The American Bazaar the rebrand reflected a real shift in direction, not just a new coat of paint: “we felt the direction of the company change from just carbon capture to something more.”
Sometime in 2025 the problem changed again. The company’s own site now frames the work as a critique of how biotechnology is organized altogether. Modalities such as antibodies, small molecules and gene therapies, it argues, are not natural categories but historical accidents, solved one at a time because that is the order in which each became technically possible. Anthrogen’s stated bet is that biology is heading toward modular design instead, where proteins are engineered and assembled the way a circuit board is specified. Carbon capture and jet fuel are no longer mentioned on the company’s homepage.
Three students, one lab
Anthrogen was founded by three Columbia University students who filed for leave to work on it full time: Ankit Singhal, named a Science Research Fellow and Named Scholar at Columbia, with years in wet and computational labs focused on catalysis and structural biology; Connor Lee, who dropped out as a sophomore after leading Columbia Robotics and researching in the university’s ROAM Lab; and Vignesh Karthik, who did machine learning and geology research at the Naval Research Laboratory and Columbia’s Lamont-Doherty Earth Observatory. A November 2024 interview with all three, conducted by an investor blog shortly after the seed round closed, has Singhal, Karthik and Lee describing themselves as “20 year olds” for whom climate change “remains at the forefront.”
By the time Odyssey shipped in October 2025, the public record had narrowed to two names. Y Combinator’s company page lists Singhal and Lee as Anthrogen’s only active founders. The Odyssey preprint on bioRxiv lists six authors, Singhal and Lee among them, but not Karthik. Ritual Capital’s investor writeup describes the company as “led by Ankit Singhal and Connor Lee.” None of this confirms that Karthik has left; it only shows that current company, platform and investor materials stop naming him where 2024 press coverage did not.
From bacteria to Odyssey
The timeline, as far as the public record shows it: Arctic Capture forms in 2023 to genetically engineer bacteria for carbon capture. It goes through Y Combinator’s Summer 2024 batch and rebrands as Anthrogen. On November 12, 2024, Axios reports a $4 million seed round co-led by Regen Ventures and BoxGroup, with Y Combinator, Wayfinder Ventures, Liquid 2 Ventures, Soma Capital, Ritual Capital, Collaborative Fund, Pioneer Fund and angels including Paul Graham, Kulveer Taggar, Todd Goldberg and Rahul Vohra’s Angel Fund. Through 2025, the company is training protein language models. On October 15, 2025, it posts “Odyssey: reconstructing evolution through emergent consensus in the global proteome” to bioRxiv, describing a family of models scaled past 102 billion parameters and 1.1 times ten to the twenty-third floating point operations, trained with a new mechanism it calls Consensus in place of self-attention. The company publishes a matching launch page and opens an API to early access. No later funding round, and no SEC Form D under either company name, has surfaced since.
What is proven, and what is still claimed
| Evidence | What the record shows | Source type |
|---|---|---|
| Odyssey preprint | Posted to bioRxiv Oct 15, 2025. Describes models scaled past 102 billion parameters, trained over 1.1x10^23 FLOPs, with a competing interest statement noting all authors are current or former Anthrogen, PBC personnel and that patents have been filed. | Public record |
| Odyssey API | Described as open for “early access.” No named customers, pricing or independent benchmark results found. | Company-stated |
| $4 million seed round | Reported closed around Nov 12 to 14, 2024, co-led by Regen Ventures and BoxGroup, with a named list of funds and angels, confirmed independently by Axios, The American Bazaar and SynBioBeta. | Independent |
| SEC Form D | No Form D found under “Anthrogen” or “Arctic Capture” in SEC EDGAR company search as of Sep 25, 2026, despite the reported round naming institutional co-leads. | Not found |
| Team size | Y Combinator’s company page lists 15 people. Tracxn, an aggregator, lists 9 employees as of Aug 26, 2026 and total funding of $500,000, well under the reported $4 million. | Differs |
| Founding team | Three co-founders named in Nov 2024 coverage (Singhal, Lee, Karthik). Y Combinator’s active founders list, the Odyssey preprint authorship and an investor writeup each name only Singhal and Lee in 2025 and 2026. | Differs |
Read plainly: the science that can be checked, the Odyssey preprint, is on the record with a dated deposit, an author list and a stated architecture. The business that can be checked, the seed round, is reported consistently by three outlets that do not work for the company. What is not on the record is any regulatory filing for that money, any named paying customer for the model, or a public accounting of why one of three founders has quietly dropped out of the company’s own materials.
What to watch
- Whether Anthrogen or a related entity ever files a Form D, or discloses a later round, that would let its funding be checked against SEC records rather than press accounts alone.
- Named commercial or research customers for the Odyssey API, beyond the “early access” description.
- Whether Vignesh Karthik remains with the company. He is named as a co-founder in 2024 reporting but is absent from Y Combinator’s active founders list, the Odyssey preprint authorship and 2025 to 2026 investor writeups.
- Peer review of the Odyssey preprint beyond its bioRxiv posting.
- Any public statement reconciling the current “post-modality biology” framing with the carbon capture and sustainable chemicals mission the company described through 2024.
In their words
“We rebranded as we felt the direction of the company change from just carbon capture to something more (producing chemicals and shifting supply chains to be inherently carbon negative) and we wanted our public facing image to reflect that.”
Ankit Singhal, CEO, The American Bazaar, Nov 14, 2024 · Independent
“The $4 million will be used to dive deeper on our R&D efforts to create novel enzymes that can capture and turn carbon dioxide into valuable fuels and other molecules like plastic, starches, and cellulose we use everyday.”
Ankit Singhal, CEO, The American Bazaar, Nov 14, 2024 · Independent
“Our goal is to produce fuels and chemicals at 80% of the cost of fossil fuel-derived products.”
Anthrogen, quoted by SynBioBeta, Nov 2024 · Company-stated
“Our biggest takeaway? Always stay focused on costs. Lack of cost-competitiveness has doomed most green startups.”
Anthrogen founders, Shared Future Fund interview, Nov 20, 2024 · Interview
“We think the next era of biology will be modular rather than modal.”
Anthrogen, company website, accessed Sep 25, 2026 · Company-stated
“It is the largest and most powerful biological model ever created.”
Anthrogen, Odyssey launch page, Oct 2025 · Company-stated
“Across various benchmarks, Odyssey achieves landmark performance for protein generation and protein structure discretization.”
Odyssey preprint abstract, bioRxiv, Oct 15, 2025 · Public record
“They’re not chasing single discoveries, they’re building the infrastructure to make discovery itself predictable.”
Ritual Capital, investor writeup, 2026 · Partner-stated
Related companies
Sources
- Public recordOdyssey: reconstructing evolution through emergent consensus in the global proteome
- IndependentAnthrogen has raised $4M to develop its AI-powered process of turning CO2 into chemicals
- IndependentColumbia students’ biotech AI startup closes $4M seed round
- IndependentAnthrogen Raises $4M to Tackle Carbon Emissions with CO2-Consuming Microbes and AI
- IndependentAnthrogen Introduces Odyssey: A 102B Parameter Protein Language Model that Replaces Attention with Consensus and Trains with Discrete Diffusion
- Independent30 Under 30 Energy & Green Tech, Anthrogen profile
- IndependentAnthrogen company profile
- InterviewAnthrogen: low-cost polymer manufacturing + CDR
- CompanyAnthrogen company website
- CompanyOdyssey launch page
- CompanyAnthrogen company profile and active founders
- PartnerThe Network Behind Biological Intelligence: The Anthrogen Story
Profile researched and written by Healthcare Discovery. Last updated September 29, 2026.
