All Benchmarks/Math

FrontierMath Tiers 1-3 (2025-02)

Older versionHigher is better

The old version 1 (290 private problems) of FrontierMath Tiers 1-3, original math problems written by mathematicians, from advanced undergraduate to early research level. It was replaced in June 2026 by version 2, which fixed errors, and is not directly comparable with version 2 scores. Scores run from 0 to 100%. Higher is better.

A newer version is available: FrontierMath Tiers 1-3 v2

Other versions:FrontierMath Tier 4 (2025-07)FrontierMath: Erdős ProblemsFrontierMath Tier 4 v2FrontierMath Tiers 1-3 v2
Top score50.0%GPT-5.4 Pro
Models tested32
Last updated2026.06.08
Model release date (newest on the right)
Best score so farHigher on the chart is better
  1. 1
    OpenAI
    GPT-5.4 Pro
    2026.03.05 · xhigh
    50.0%
  2. 2
    OpenAI
    GPT-5.4
    2026.03.05 · xhigh
    47.6%
  3. 3
    Anthropic
    Claude Opus 4.8
    2026.05.27 · max
    47.2%
  4. 4
    Anthropic
    Claude Opus 4.7
    2026.04.16 · xhigh
    43.8%
  5. 5
    Anthropic
    Claude Opus 4.6
    2026.02.04 · max
    40.7%
  6. 6
    Meta
    Muse Spark
    2026.04.08
    39.0%
  7. 7
    Google
    Gemini 3.5 Flash
    2026.05.19 · high
    39.0%
  8. 7
    Moonshot AI
    Kimi K2.6
    2026.04.20
    39.0%
  9. 9
    Google
    Gemini 3.1 Pro
    2026.02.19
    36.9%
  10. 10
    Google
    Gemini 3 Flash
    2025.12.17
    35.6%

The "harness" is the agent program the AI used to carry out the task. The same model can score very differently depending on its harness and reasoning effort.