Wikibench

CountBench

reasoning

Multilingual: no

CountBench evaluates object counting capabilities in visual understanding.

Methodology

Imported from llm-stats public benchmark metadata. Modality: image. Max score: 100. Categories: reasoning, spatial_reasoning, vision. Language: en. Verified by llm-stats: no.

Leaderboard

Showing 6 of 6 results

  1. Qwen3.5-27B self-reported llm-stats
    97.8%
  2. Qwen3.5-35B-A3B self-reported llm-stats
    97.8%
  3. Qwen3.6-27B self-reported llm-stats
    97.8%
  4. Qwen3.6 Plus self-reported llm-stats
    97.6%
  5. Qwen3.5-122B-A10B self-reported llm-stats
    97.0%
  6. Qwen3 VL 235B A22B Thinking self-reported llm-stats
    93.7%