Share your thoughts, 1 month free Claude Pro on usSee more

Counting on CountBench (test)

91.5Accuracy

Dr. Seg

Updated 1mo ago

Evaluation Results

Method	Links
Dr. Seg 2026.02		91.5
Claude-4-Sonnet 2026.06		90.2
Gemini2.5-Pro 2026.06		89.7
Visual Para-Thinker++ 2026.06		88.3
VisionReasoner 2026.02		87.6
Qwen2.5-VL 2026.06		86.7
GPT-4o 2026.06		85.7
Visual Para-Thinker 2026.06		85.4
GPT5-mini 2026.06		84.4
GRPO 2026.06		84.2
Sequential (SFT) 2026.06		83.7
Majority voting@4 2026.06		83.2
Qwen2.5-VL 2026.06		83.1
Qwen2-VL 2026.02		76.5
Qwen2.5-VL 2026.02		76
Visual Para-Thinker++ 2026.06		74.8
Visual Para-Thinker 2026.06		72.6
GRPO 2026.06		71.2
Sequential (SFT) 2026.06		70.4
Majority voting@4 2026.06		69.8
Qwen2.5-VL 2026.06		69.2