we build
foundation models
for biology.
Living Models is building the AI infrastructure to understand living systems — starting with the plant kingdom.
a family built to find
biological signal
BOTANIC-1 sets a new state of the art on plant-genomics benchmarks. Trained across 320 species, the family also leads our new, harder benchmark built from experimentally validated causal variants.
3B
Parameters in our BOTANIC-1 flagship
49.9%
Studies where the known causal variant lands in the top 1%
128 kbp
Maximum genomic context
models that read
plant DNA directly
Most AI systems can only reach genomics through tools designed by humans. BOTANIC-1 learns directly from plant DNA, one nucleotide at a time.
The model learn what a genome encodes: gene boundaries, splice sites, regulatory motivfs, and how a single mutation changes functions. That understanding is what turns a sequence into something a biologist can act one.
All 24 models as a table
| Model | Organisation | Parameters | Base pairs | Score |
|---|---|---|---|---|
| BOTANIC-1 XL | Living Models | 3.2B | 314.6B | 0.769 |
| BOTANIC-1 L | Living Models | 2.1B | 314.6B | 0.768 |
| BOTANIC-1 M | Living Models | 688M | 314.6B | 0.765 |
| BOTANIC-1 S | Living Models | 318M | 314.6B | 0.758 |
| PlantCAD2-L | Cornell University | 694M | 4.03T | 0.756 |
| PlantCaduceus | Cornell University | 225M | 251.7B | 0.752 |
| PlantCAD2-S | Cornell University | 88M | 4.03T | 0.742 |
| PlantCAD2-M | Cornell University | 311M | 4.03T | 0.733 |
| GPN | UC Berkeley | 66M | 157.3B | 0.731 |
| Carbon-3B | Hugging Face | 3.5B | 1.05T | 0.731 |
| Carbon-8B | Hugging Face | 8.3B | 1.05T | 0.727 |
| NTv3 (pre) | InstaDeep | 652M | 10.85T | 0.725 |
| Carbon-500M | Hugging Face | 512M | 600B | 0.709 |
| PlantBiMoE | Huazhong Agric. Univ. | 116M | 241.3B | 0.704 |
| Evo 2-7B | Arc Institute | 7B | 2.4T | 0.695 |
| NTv3 | InstaDeep | 680M | 12.1T | 0.690 |
| BOTANIC-0 L | Living Models | 991M | 226.1B | 0.681 |
| Evo 2-20B | Arc Institute | 20B | 9.3T | 0.678 |
| Evo 2-1B | Arc Institute | 1B | 1T | 0.669 |
| PlantGFM | Huazhong Agric. Univ. | 220M | 430.6B | 0.650 |
| BOTANIC-0 M | Living Models | 260M | 226.1B | 0.643 |
| Evo 2-40B | Arc Institute | 40B | 9.3T | 0.629 |
| AgroNT | InstaDeep | 1B | 472.5B | 0.575 |
| BOTANIC-0 S | Living Models | 114M | 226.1B | 0.548 |
the discovery pipeline is broken
Developing a single crop trait takes 8+ years. The $60B crop protection market relies on pipelines built for a pre-genomics era. Foundation models can change that.
Crop protection market
Years to develop a new trait
Yield losses from climate change
from 47,492 variants
to the one that matters
Elucidating a trait can take years and thousands of plants. Its genetic architecture has to be understood, and the experiments that resolve it at the right genetic scale are slow to produce. In a published melon flower sex-determination study, BOTANIC-1 ranked the causal mutation in CmEIN3 first out of 3,061 candidates.
This is not one lucky story. Across 545 experimentally validated loci in 14 species, BOTANIC-1 places the causal variant in the top 1% of candidates for half of them. Genetics tells you where to look. BOTANIC-1 tells you what to test first.
47,492
variants across the melon genome
3,061
on chromosome 2, where the correlation signal points
#1
the causal base, top score from BOTANIC-1
a research lab
that runs as a loop
BOTANIC-1 is the first model out of our Model Factory. Data curation, training, evaluation, and debugging run as one orchestrated loop, with researchers setting the questions and judging the results.
bring us a trait
Narrowing a mapped region, choosing which edits to make, or predicting traits in your own populations. We run a research evaluation of BOTANIC-1 next to your current tools, on your data, and measure what it adds.
hello@livingmodels.ai
Paris & Berkeley
PhDs and engineers from Huawei Noah's Ark Lab, Owkin, Datadog, UMass Chan Zuckerberg Medical School, Institut Pasteur, Polytechnique.
We're building something ambitious. Come help.
Backed by
Pre-Seed backing from leading investors in frontier AI and deep tech.