Wikibench

MultiChallenge (o3-mini grader)

reasoning official site →

Multilingual: no

A realistic multi-turn conversation evaluation benchmark that challenges frontier LLMs across four key areas: instruction retention, inference memory, reliable versioned editing, and self-coherence. Despite near-perfect scores on existing benchmarks, frontier models achieve less than 50% accuracy on MultiChallenge.

Methodology

Imported from llm-stats public benchmark metadata. Modality: text. Max score: 1. Categories: reasoning, language. Language: en. Multilingual: no. Verified by llm-stats: no.

Leaderboard

Showing 7 of 7 results

  1. GPT-5 self-reported llm-stats
    69.6%
  2. o3-mini self-reported llm-stats
    50.2%
  3. GPT-4.5 self-reported llm-stats
    50.1%
  4. GPT-4.1 self-reported llm-stats
    46.2%
  5. GPT-4.1 mini self-reported llm-stats
    42.2%
  6. GPT-4o self-reported llm-stats
    39.9%
  7. GPT-4.1 nano self-reported llm-stats
    31.1%