A new blog post on Hugging Face describes training mRNA language models using data from 25 species. The authors report a total training cost of $165, a figure that stands out because genomic language models are often assumed to require substantial compute resources.
The post emphasizes the low cost relative to typical large-scale language model training, suggesting that such models can be developed with limited budgets. It focuses on the training setup and the breadth of species included.
As a single source, the post does not offer comparisons with other models or benchmarks. It also does not include details on model accuracy or downstream biological tasks, so the practical impact of these low-cost models remains unclear.