Methodology
How every number on llmrun is produced.
3 reference pages · updated Oct 24, 2026
Articles
How the llmrun Composite Score is built
How llmrun combines public LLM benchmarks into one open-ended Composite Score with Item Response Theory, a 90% range and per-skill scores.
Oct 3 · 8 min read
Which benchmarks llmrun uses
A plain-English guide to the LLM benchmarks llmrun aggregates: what each one tests, how to read the scores, and open vs proprietary models.
Jun 7 · 7 min read
How llmrun determines model requirements
How llmrun decides whether a GPU can run an LLM: VRAM, quantization, decode and prefill speed, time to first token and the S–F Fit scale.
Mar 15 · 9 min read