Charts
Starting points for common questions. Every chart is interactive: change either axis, colour by lab or region, hover for details, or click a point to open the model.
Rankings
Top 15 models by the index or any single benchmark.
Top 15 by Small LM Index
Efficiency: intelligence per active parameter
Active parameters per token drive inference cost and speed. MoE models sit far to the left of dense models of similar quality.
Frontier: nothing beats these on both axes (labelled)87 plotted · 52 have no data for these axes
Progress over time
Small LM Index by release date: how fast small open models are improving.
87 plotted · 52 have no data for these axes
Agentic coding vs size
SWE-bench Verified against total parameters. Harness and scaffolding differ between labs.
Frontier: nothing beats these on both axes (labelled)27 plotted · 112 have no data for these axes
Popularity vs intelligence
Hugging Face downloads in the last 30 days (main repository only, quantized copies not counted) against the index.
Frontier: nothing beats these on both axes (labelled)85 plotted · 54 have no data for these axes