Wikibench

Multi-Challenge

reasoning official site →

Multilingual: no

MultiChallenge is a realistic multi-turn conversation evaluation benchmark that challenges frontier LLMs across four key categories: instruction retention (maintaining instructions throughout conversations), inference memory (recalling and connecting details from previous turns), reliable versioned editing (adapting to evolving instructions during collaborative editing), and self-coherence (avoiding contradictions in responses). The benchmark evaluates models on sustained, contextually complex dialogues across diverse topics including travel planning, technical documentation, and professional communication.

Methodology

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

Leaderboard

Showing 20 of 22 results

  1. Nova 2 Pro self-reported llm-stats
    77.7%
  2. Nova 2 Lite self-reported llm-stats
    76.6%
  3. Nova 2 Omni self-reported llm-stats
    75.5%
  4. GPT-5 self-reported llm-stats
    69.6%
  5. Qwen3.5-397B-A17B self-reported llm-stats
    67.6%
  6. Qwen3.5-122B-A10B self-reported llm-stats
    61.5%
  7. Qwen3.5-27B self-reported llm-stats
    60.8%
  8. o3 self-reported llm-stats
    60.4%
  9. Qwen3.5-35B-A3B self-reported llm-stats
    60.0%
  10. Nemotron 3 Super (120B A12B) self-reported llm-stats
    55.2%
  11. Kimi K2 Instruct self-reported llm-stats
    54.1%
  12. Kimi K2-Instruct-0905 self-reported llm-stats
    54.1%
  13. MAI-Thinking-1 self-reported llm-stats
    53.0%
  14. MiniMax M1 40K self-reported llm-stats
    44.7%
  15. MiniMax M1 80K self-reported llm-stats
    44.7%
  16. GPT-4.5 self-reported llm-stats
    43.8%
  17. GPT-4o self-reported llm-stats
    40.3%
  18. o3-mini self-reported llm-stats
    39.9%
  19. Nemotron 3 Nano (30B A3B) self-reported llm-stats
    38.5%
  20. GPT-4.1 self-reported llm-stats
    38.3%