Skip to content

ZeroBench

A multi-step visual reasoning benchmark with pass@5 reporting and optional tool use.

8 models

Higher is better.

  1. 1Muse Spark OpenMeta50.0
  2. 2GPT-5.4OpenAI41.0
  3. 3Muse SparkMeta33.0
  4. 4Gemini 3.1 ProGoogle29.0
  5. 5Qwen3.8 MaxAlibaba24.0
  6. 6Kimi K3Moonshot AI23.0
  7. 7dots3-note PreviewDots Studio19.0
  8. 8Seed 2.1 ProByteDance18.0
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Multi-step visual reasoning
Tasks
100 visual reasoning questions
Authors
Meta AI (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.