The Complete Benchmark

MARC expands on the ARC challenge with multimodal and reasoning-heavy tasks, creating the most comprehensive general intelligence benchmark to date.

Core ARC-Style Tasks

40 puzzles

  • Pattern recognition and transformations
  • Grid-based reasoning challenges
  • Abstraction without domain-specific hacks
  • Novel problem settings to prevent memorization

Math & Logical Reasoning

10 word problems

  • Algebra and probability challenges
  • Logic puzzles requiring deep reasoning
  • Numerical reasoning tasks
  • Novel contexts to prevent pattern memorization

Audio Understanding

10 audio challenges

  • Spoken instruction comprehension
  • Overlapping sounds analysis
  • Rhythm and melody pattern matching
  • Environmental sound reasoning

Visual Reasoning

10 image tasks

  • Odd-object identification
  • Visual transformation analysis
  • Pattern recognition in sequences
  • Spatial reasoning challenges

Retrieval-Augmented Generation

10 RAG questions

  • Information retrieval from provided corpus
  • Combining retrieval with reasoning
  • Comparative analysis tasks
  • Multi-document synthesis

Video Understanding

10 video clips

  • Next frame/event prediction
  • Temporal sequence reasoning
  • Causal relationship understanding
  • Dynamic scene analysis

Why MARC Will Become the Standard

🎯 Multimodal & Multi-Reasoning

Unlike ARC alone, MARC tests across all core modalities that general AI must handle

🚫 Prevents Gaming

Forces models to generalize across modalities instead of becoming benchmark-specialists

🎨 Comprehensive Coverage

Tests abstraction, math, language, vision, audio, retrieval, and video understanding

⚖️ Fair & Balanced

Each subtask contributes equally with partial credit for reasoning tasks

Ready to Test Your AI?

Join the competition and benchmark your model against the most comprehensive general intelligence test

Register Your Team Download Dataset View Leaderboard