GRM-Kerlin-3b
GRM-Kerlin is our specialized general-purpose puzzle reasoning model, trained to solve puzzles of any type.
Highlights
- Efficient puzzle solving: capable of solving difficult puzzles with ease
- Thinks before acting: reasons through the task's logic and recognizes its own errors
- Compact intelligence: logical reasoning at the level of large models, with only 3 billion parameters
Orion Research - 2026