GeneICL: 4.2M-parameter model matches foundation models 387x larger
anshulkundaje · x · 2026-10-08
Theus Research reports that GeneICL achieves the best overall rank across evaluated methods for gene function prediction. With only 4.2M parameters it matches or outperforms much larger foundation models (up to 387x fewer parameters) and runs within seconds on a laptop CPU — a sign that small specialized models can beat general-purpose FMs on domain tasks.
Related event: GeneICL: Lightweight Transcriptomics Model Beats Larger Baselines(2 posts)→
More from Models
- Users complain OpenAI is silently swapping models mid-chat even on the Pro plan — Slow_Ad1827 · 2026-10-09
- GPT-6 reportedly rolling out with Intelligent UI — answers embed interactive tools — winer666 · 2026-10-09
- Anthropic's push for "clean" training data criticized for ignoring real human behavior — ssh4net · 2026-10-09
- COLM poster: behaviors thinking models amplify aren't the ones that drive good outcomes — Jeande_d · 2026-10-09
- EngramEdit: near-perfect fact editing in LLMs via conditional memory, 3x baseline CoT accuracy — teortaxesTex · 2026-10-09
- ThursdAI: OpenAI drops 722 math papers, Haiku 5.5 hits 10 cents, Arena raises $200M — altryne · 2026-10-09