Atlas

Benchmarks

← All benchmarks

Roboflow Vision Evals - Object Counting

Multimodal · 2026-07-22

The Roboflow Vision Evals counting task measures exact-match accuracy on 74 image questions that require counting visible objects.

Top models (higher is better)

ModelScore
Gemini 3.6 Flash82.4
Gemini 3.5 Flash81.1
Muse Spark 1.175.7
GPT-5.6 Sol73.0
Gemini 3.1 Pro Preview71.6
Gemini 3 Flash Preview67.6
GPT-5.6 Terra67.6
GPT-5.6 Luna66.2
GPT-5.564.9
Claude Fable 563.5
GPT-5.4 Mini60.8
Sonnet 556.8
Qwen3.5-27B54.0
Opus 4.852.7
Gemini 2.5 Pro52.7
Gemini 3.5 Flash-Lite52.7
Kimi K2.650.0
Qwen3.7-Plus50.0
GLM 5V Turbo48.6
Qwen3-VL-235B-A22B-Instruct47.3
Kimi K346.0
Loading Atlas data…