Researchers developed RIBOSPAN, a 1.61-billion-parameter RNA foundation model using dense bidirectional self-attention, single-nucleotide tokenization, and attention-isolated sequence packing to process complete RNA transcripts up to 10,240 nucleotides. The model was evaluated through nucleotide reconstruction tasks and specialized benchmarks measuring contextual representation quality across extended sequences. The native long-context architecture maintains strong reconstruction fidelity while enabling high-resolution full-transcript modeling, with applications including mRNA generation and protein-preserving codon optimization.
