LLM Benchmarks & Leaderboards

Each benchmark below ranks large language models on a specific task — and pairs every score with the model's cost per million tokens, so you can see which model is the best value, not just the highest score.

LMArena (Chatbot Arena) Elo

228 models

human-preference

View leaderboard →

Epoch Capabilities Index

194 models

composite

View leaderboard →

GPQA Diamond

169 models

reasoning

View leaderboard →

LMArena WebDev Elo

115 models

coding

View leaderboard →

LMArena Vision Elo

113 models

vision

View leaderboard →

LiveBench

87 models

reasoning/coding

View leaderboard →

FrontierMath

60 models

math

View leaderboard →

SWE-bench Verified

52 models

coding

View leaderboard →

Aider Polyglot

40 models

coding

View leaderboard →

SWE-bench Multilingual

13 models

coding

View leaderboard →

SWE-bench Multimodal

8 models

coding

View leaderboard →

Decision-model benchmarks

Decision models answer typed questions with calibrated probabilities instead of generating text. These boards rank them on accuracy, with calibration (ECE) and latency alongside. See all decision models.

Jev Decision Index

189 models

decision

View leaderboard →

Benchmark Heaven JevBench (Capability)

163 models

decision

View leaderboard →

DecisionBench

42 models

decision

View leaderboard →

Jev Decision Index — Vision

33 models

decision-vision

View leaderboard →

Typed Decisions (agreement with teacher)

32 models

decision-synthetic

View leaderboard →

JevBench coherence

21 models

decision-diagnostic

View leaderboard →

PT-BR typed decisions

16 models

decision-language

View leaderboard →

Enterprise Decision Arena (provisional)

12 models

decision

View leaderboard →

Typed Decision Leaderboard (answer verifiers)

11 models

decision-diagnostic

View leaderboard →

System One benchmark

Decision models, not chat models: TypeSafe's Jev and every open alternative (LAYA, Von, GLiNER2, GLiClass) measured on 2,123 judgments through one harness — accuracy, calibration, latency.

View the System One leaderboard →

Looking for the underlying datasets? Browse our benchmark datasets (MMLU, GPQA, HumanEval and more).