Chart ranks open models by score-per-parameter efficiency
This roundup defines a model's efficiency as its benchmark score divided by its number of , then keeps only the models sitting on the efficiency frontier () — meaning every other model was dropped because some listed model matches or beats it while using fewer or equal . The benchmark scores come from 's aggregate benchmark, which is described as imperfect but a generally good overall indicator.
Key points
- Efficiency is defined as benchmark score divided by
- Only models on the efficiency frontier are kept; dominated models are removed
- Benchmark scores are sourced from 's aggregate benchmark