Botanic 1

BOTANIC-1 is a genetic language model (GLM), described as a plant genomic world model, built by the Paris-based AI biology lab Living Models. It is pre-trained across 320 plant species and is meant to capture evolutionary constraints in plant genomes. Google DeepMind features it on its Gemma "Gemmaverse" pages as an example of Gemma 4 being used in a domain-specific pipeline.

In the featured pipeline, Gemma 4 E4B acts as an agentic orchestrator. It translates natural-language questions into structured queries against BOTANIC-1's inference endpoints, and it also writes bioinformatics pipelines and runs tool calls. BOTANIC-1 scores candidate mutations. In a melon test, the CmEIN3 gene (flower sex) was narrowed from 47,492 genome-wide variants to 3,061 regional candidates. A creator summary says the model ranked the correct mutation first among about 2,500 candidates 90% of the time. The source page compares this against a classical-tools setup (bcftools, Ensembl VEP). That setup in its best runs left two identically scored coding mutations.

Living Models runs Gemma 4 E4B locally via Ollama on a single NVIDIA L4 GPU, so proprietary genomes stay on the user's own infrastructure. The provided material does not describe public weights, pricing or a public API for BOTANIC-1. It appears to be a research and partner system rather than a general release.