Living Models has exited stealth mode with $7 million in seed funding to accelerate the development of AI foundation models for plant genomics, technology it claims could meaningfully shorten crop breeding timelines and expand access to genomic selection far beyond today’s largest seed companies.
The round was co‑led by Asterion Ventures and The Galion Project, joined by Kima Ventures, STATION F and several strategic investors. The Franco‑American company also released its first technical report, published research findings, and made its BOTANIC foundation model openly available on Hugging Face for the scientific community.
A new class of genomic AI trained across 1,600 plant genomes
Living Models’ flagship technology, BOTANIC, is a family of transformer models trained on genomic sequences from 43 plant species. With up to 1 billion parameters, BOTANIC reportedly matches the performance of state‑of‑the‑art genomic models across 22 benchmark tasks — despite being trained on just eight NVIDIA H100 GPUs.
The fresh capital will fund a major compute upgrade: a dedicated 120‑GPU NVIDIA B200 cluster, which the company says will enable significantly larger models, improved predictive accuracy and expansion beyond plants into broader genomic domains.
CEO and co‑founder Cyril Véran told AgTechNavigator that the core motivation is simple: crop breeding still runs on an 8-12‑year cycle, too slow for accelerating climatic volatility.
“A breeder’s workflow is largely defined by what they can afford to phenotype,” he said. “Genomic selection improved this, but those models are narrow, crop‑specific and data‑hungry. Smaller programmes, new crops and under‑characterised traits get left behind.”
From correlation to biological signal: ‘the upgrade from prediction to understanding’
What makes BOTANIC different, Véran argued, is the ability to transfer biological structure across species and traits.
“Because it was pre‑trained on 1,600 genomes across the plant kingdom, it encodes deep biological structure that transfers across species,” he said. “Breeders can run higher‑confidence early‑stage selection on smaller panels, reduce field cycles and expand into traits that were previously too expensive to screen at scale.”
More importantly, he stressed, BOTANIC moves beyond classical statistical pattern‑matching.
“Standard genomic selection treats markers as exchangeable statistical surrogates. BOTANIC learns the functional genomic signatures associated with a trait. Breeders aren’t just getting a better number; they’re getting a more biologically grounded signal.”
This distinction, he said, is what shifts the tool from prediction to understanding.
Potential to cut 1-4 years from the breeding cycle
Asked how much this could accelerate commercial product development, Véran was cautious but optimistic.
Academic literature, he said, shows improved early‑stage genomic prediction can remove one to three field cycles – often 1-4 years. “Faster cycles mean climate-adapted varieties reaching commercial release sooner and with better-characterized performance across environments.”
He believes traits with strong genetic architecture and available training data, including yield stability, drought tolerance, and disease resistance, are expected to show the earliest gains.
In yield stability, BOTANIC’s embeddings reportedly begin to identify genomic regions contributing to performance across environments, converting a historical “black box” prediction into mechanistically actionable insight.
Disease resistance may prove even more compelling, he believes. “Resistance loci often transfer across species. BOTANIC has already seen how resistance is encoded across the plant kingdom, which makes it far more sensitive to relevant signatures in any individual crop system.”
Accessible to more than the top five seed companies
Véran argues the larger significance is accessibility. Today, high‑accuracy genomic prediction is typically limited to organisations with vast internal data infrastructure.
“BOTANIC delivers strong predictions even where internal datasets internal data is limited. The same platform serves a global seed company’s elite germplasm program and a regional breeder working with locally adapted varieties, calibrated to what each brings to it.”
Initial commercial traction is emerging in North America and Europe, supported by academic partnerships with INRAE, Paris‑Saclay Plant Science Institute, CRAG Barcelona, Aarhus University and the University of Florida. Discussions are underway with a top 5 seed company, Véran claimed, alongside several regional breeders.
But the company’s longer‑term ambition focuses on underserved crop systems such as sorghum, cassava and millet.
“The platform was built so a smallholder‑focused breeding programme in sub‑Saharan Africa can benefit as much as a commercial programme in Iowa. That’s not a stretch goal. It’s the point.”
Licensing model aims to keep breeder data private
Living Models will commercialise BOTANIC through a licensing model, allowing seed companies and research institutions to fine‑tune the AI within their own secure environments.
“We don’t touch their data; they retain full ownership,” Véran said. “The problem isn’t that the data doesn’t exist — it’s that current modelling tools can’t extract its full value. BOTANIC is built to change that equation.”
The company is co‑developing integrations directly with customers to ensure BOTANIC fits seamlessly into existing R&D workflows rather than disrupting them.
“Biology is an information problem at every scale, from a single cell to an entire ecosystem,” Leonard Strouk, CTO and co-founder The genomic data exists across many domains; what’s been missing is a model architecture capable of learning from it at scale.
“Every living thing on Earth runs on the same programming language: DNA codes for RNA codes for proteins codes for phenotype,” added Bertrand Gakière, VP Biology. “We’re not building another chatbot. We’re building a model that can read and interpret that code, which is infinitely more useful than predicting the next word in a sentence.”
