Skip to content
LLM Atlas

Benchmarks

36 of 80 benchmarks here are saturated or deprecated — they no longer separate strong models from weak ones, yet scores on them are still quoted. Rows are capabilities, columns are how much a benchmark still discriminates. fetched 2026-10-19

Benchmark metadata from benchwiki, which researches and maintains these health records. This site mirrors them to join scores to them; the records are theirs.

!
high contamination risk
◇
models on this site have scores on it
Benchmarks by capability and status. Rows are capabilities, columns are how much the benchmark still discriminates between models.
Capabilityactivenearing saturationsaturateddeprecated
Agentic tool use5 benchmarks·
Coding16 benchmarks
Instruction following6 benchmarks·
Knowledge6 benchmarks·
Long context11 benchmarks·
Maths12 benchmarks
Multimodal5 benchmarks·
Reasoning10 benchmarks·
Safety and alignment9 benchmarks