Skip to main content

Benchmark profile

MATH-500 Problem Set (MATH-500)

A curated subset of 500 problems from the MATH dataset, covering algebra, counting and probability, geometry, intermediate algebra, number theory, prealgebra, and precalculus.

Data verified

Benchmark score on MATH-500 — July 20, 2026

BenchLM mirrors the published score view for MATH-500. MiniCPM5-1B leads the public snapshot at 91.6% , followed by LFM2.5-8B-A1B (88.8%). BenchLM does not use these results to rank models overall.

2 modelsMathStaleDisplay onlyUpdated July 20, 2026

Benchmark score table (2 models)

Score
1
MiniCPM5-1BOpenBMB · Open weight
91.6%
2
LFM2.5-8B-A1BLiquidAI · Open weight
88.8%

About MATH-500

Year

2021

Tasks

500 problems

Format

Free-form mathematical answers

Difficulty

High school to undergraduate

MATH-500 is one of the most widely cited math benchmarks. It is nearing saturation with top reasoning models scoring 96-99%, making it less useful for differentiating frontier models but still a standard baseline.

BenchLM freshness & provenance

Version

MATH-500 2021

Refresh cadence

Static

Staleness state

Stale

Question availability

Public benchmark set

StaleDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

FAQ

What does MATH-500 measure?

A curated subset of 500 problems from the MATH dataset, covering algebra, counting and probability, geometry, intermediate algebra, number theory, prealgebra, and precalculus.

Which model scores highest on MATH-500?

MiniCPM5-1B by OpenBMB currently leads with a score of 91.6% on MATH-500.

How many models are evaluated on MATH-500?

2 AI models have been evaluated on MATH-500 on BenchLM.

Compare Top Models on MATH-500

Last updated: July 20, 2026 · BenchLM version MATH-500 2021

Choose a model with this week’s evidence

Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.

One email each week. Unsubscribe anytime.