LLM Ranking: 257 Language Models on One Scale

Mon, 12 Oct 2026 14:00:00 +0200

Single benchmarks disagree: one model leads in mathematics, another in coding, a third in long tasks. A ranking needs one scale. We use the Epoch Capabilities Index, which combines the results of more than 50 benchmarks into one number per model — an index, not a percentage, scaled so that Claude 3.5 Sonnet stands at 130 and GPT-5 at 150.

The ranking covers the 257 models for which the index exists. We did not test any model ourselves and changed no value; our part is the ranking views below. The margins at the top are small: differences of one or two index points are within the uncertainty of the index.

LLM Benchmarks 2019–2026: The Records on 16 Leaderboards

Mon, 12 Oct 2026 13:00:00 +0200

A benchmark score means little on its own. It starts to mean something when you see how fast it moved: which model set the record, when, and how long the test lasted before the next one had to be invented.

This page keeps that record for 16 benchmarks, from the GPT-2 era to today. For each one you see every model that beat all earlier models at the time of its release, and the leaders as of now. The numbers come from the open benchmark archive of Epoch AI — 2,621 results in the 16 benchmarks shown here. We did not test any model ourselves and we changed no value; our part is the selection, the record lines and the comparison over time.

LLM Database — ABXK

LLM Ranking: 257 Language Models on One Scale

LLM Benchmarks 2019–2026: The Records on 16 Leaderboards