BOTANIC-1's 3.2 Billion Parameters Aim to Decode Crop DNA

A transformer model sees a string of nucleotides from a soybean genome. Its job is not to predict the next word, but the function of a gene, the location of a regulatory switch, or the link between sequence and drought tolerance. This is the core task for BOTANIC, the plant genomic language model from Paris-based Living Models. The company, which emerged from stealth in March with a $7 million seed round, is applying the architecture that reshaped natural language processing to what it calls the fundamental language of life: DNA [Tech.eu, March 2026].

Its first commercial wedge is agricultural genomics, a field with deep data troves and a pressing need to accelerate the decade-long cycle of breeding new crops. Living Models is not building another drug discovery platform. Instead, it is training its models on public and proprietary plant sequence data, offering them as an open-weight foundation for seed companies to fine-tune for their own pipelines [Perplexity Sonar Pro Brief, retrieved 2026]. As of May, the company was pre-revenue and speaking with 10 large seed companies as potential development partners [Diginomica, May 2026].

The technical wedge in plant biology

Living Models is betting that a scaled, domain-specific foundation model can outperform narrower tools. The initial BOTANIC model was trained on data from 43 plant species [Crunchbase News, April 2026]. The company has since released BOTANIC-1, a larger model family trained on sequences from 320 species, with a flagship model boasting 3.2 billion parameters and a genomic context window of 128 kilobase pairs [Living Models, retrieved 2026]. These scales are not arbitrary. A longer context allows the model to analyze larger stretches of DNA, capturing interactions between distant regulatory elements and genes, which is critical for understanding complex traits.

The models are openly available on Hugging Face, a strategic choice for a pre-commercial company. It lowers the barrier for research evaluation and allows potential partners to test performance on their own data before committing to a commercial relationship. A February 2026 preprint on bioRxiv reported that the earlier BOTANIC-0 models (spanning 100 million to 1 billion parameters) achieved competitive performance with state-of-the-art foundation models in both zero-shot settings and after fine-tuning [bioRxiv, February 2026]. This open-weight approach is intended to prove utility in corporate R&D environments before a sales motion begins [CB Insights, retrieved 2026].

A team built for the intersection

The founding trio is engineered to bridge AI research and plant biology. CEO Cyril Véran is described as a serial entrepreneur at the intersection of AI and plant biology [Living Models, retrieved 2026]. CTO Léonard Strouk is a biochemist turned AI engineer, trained at École Normale Supérieure and UC Berkeley, who previously founded a generative AI company [Living Models, retrieved 2026] [Crunchbase Person Profile, retrieved 2026]. The scientific anchor is CSO Bertrand Gakière, an associate professor in plant biology at Université Paris-Saclay with over 75 publications and multiple trait patents [Living Models, retrieved 2026].

The technical team draws from labs and companies including Huawei Noah’s Ark Lab, Owkin, Datadog, and Mila, suggesting a blend of deep learning research and production engineering experience [storiesout.com, retrieved 2026].

BOTANIC-0 (Feb 2026) | 43 | plant species
BOTANIC-1 (2026) | 320 | plant species

The seed round and strategic backers

The $7 million seed round closed in March 2026 was led by Artesian VC and included a notably strategic mix of investors [Tech.eu, March 2026]. The cap table features UC Berkeley’s investment arm and its SkyDeck accelerator, Juniper, Asterion Ventures, and Pascual Innoventures, which is the corporate venture arm of dairy giant Grupo Pascual. GrainCorp, a major Australian agricultural business, also participated through Artesian [Tech.eu, March 2026] [Crunchbase News, April 2026]. This blend provides more than capital. It offers academic credibility, potential enterprise pathways, and direct links to agricultural supply chains.

The funding is earmarked for scaling the model training infrastructure and expanding the team. The company’s Paris-Berkeley axis is deliberate, tapping into France’s strong academic base in mathematics and AI, and the Bay Area’s venture and commercial biotechnology networks.

Navigating a pre-revenue path to market

The company’s current phase presents its most defined challenge. It has built a technically ambitious model and secured quality backing, but it has not yet converted a development partner into a paying customer. The agricultural biotech market is dominated by large, established seed companies and a handful of agribusiness giants. These are sophisticated buyers with internal data science teams and long, risk-averse evaluation cycles.

Living Models’ strategy appears to be a classic wedge: prove technical superiority and utility through open access, then monetize through enterprise licenses for fine-tuned, proprietary versions of the model or through a managed API service. The active discussions with 10 seed companies are a critical leading indicator. Success will be defined by converting one of those dialogues into a paid pilot project that demonstrates a measurable acceleration in a trait discovery pipeline.

The scale-up questions

Technically, the jump from 43 to 320 species for BOTANIC-1 is significant, but the real test is functional accuracy on novel, commercially valuable predictions. The model’s architecture must handle the noise and complexity of real-world genomic data, which includes structural variations, epigenetic marks, and incomplete annotations. The 128 kbp context is a good start, but some regulatory interactions in plants can span much longer genomic distances.

From an infrastructure perspective, serving inference for models of this size to multiple enterprise clients, each with their own fine-tuned variants and data privacy requirements, is a non-trivial engineering challenge. It requires a robust, scalable serving layer that the company has not yet had to build for public Hugging Face downloads.

The commercial risk is timeline. Seed companies operate on long breeding cycles. Proving that the AI model can shave meaningful time or cost off that process requires validation in greenhouses and field trials, not just on a benchmark dataset. This sales cycle could stretch well beyond the 18-24 month runway a typical $7 million seed round provides, potentially necessitating a bridge or Series A round before material revenue is secured.

The model’s performance on polygenic traits, which are controlled by many genes and heavily influenced by the environment, will be the ultimate benchmark. If BOTANIC can consistently identify candidate genes for complex attributes like heat tolerance or nitrogen use efficiency, it will have found its product-market fit. If its predictions remain confined to simpler, monogenic traits, it becomes a useful tool rather than a transformative platform. The next twelve months will be about moving from technical preprint to a signed case study with a partner willing to attach its name to the results.

Sources

1. [Tech.eu, March 2026] Decoding DNA with AI: Living Models emerges from stealth with $7m | https://tech.eu/2026/03/11/decoding-dna-with-ai-living-models-emerges-from-stealth-with-7m/
2. [Crunchbase News, April 2026] 5 Interesting Startup Deals You May Have Missed: A Credit … | https://news.crunchbase.com/venture/interesting-startup-deals-mineral-rights-flying-ferry-ai-clean-tech/
3. [Diginomica, May 2026] Being best is good, being useful is better: how Living Models’ BOTANIC AI sets out to tackle climate change. | https://diginomica.com/being-best-good-being-useful-better-how-living-models-botanic-ai-sets-out-tackle-climate-change
4. [Living Models, retrieved 2026] Living Models, AI for Living Systems | https://www.livingmodels.ai/
5. [Perplexity Sonar Pro Brief, retrieved 2026] Living Models briefing
6. [bioRxiv, February 2026] BOTANIC-0: Plant Genomic Language Models | https://www.biorxiv.org/content/10.1101/2026.02.XX.XXXXXXv1
7. [CB Insights, retrieved 2026] Living Models company profile
8. [storiesout.com, retrieved 2026] Living Models team profile
9. [Crunchbase Person Profile, retrieved 2026] Léonard Strouk profile
10. [Hugging Face, retrieved 2026] BOTANIC model repository
11. [AlphaSignal, retrieved 2026] BOTANIC-1 announcement
12. [explainx.ai Blog, 2026] BOTANIC-1 model description