Atlas

Benchmarks

← All benchmarks

VitaBench

Professional Work · 2025-09-30

VitaBench evaluates agents on 400 interactive daily-life tasks spanning food delivery, in-store consumption, online travel services, and cross-scenario workflows with 66 tools. This aggregate is the macro-average task success rate across the four 100-task domains.

Top models (higher is better)

ModelScore
Qwen3.7-Max61.2
Macaron-V1-Venti60.0
MiniMax M356.8
Opus 4.856.5
GLM-5.255.8
GPT-5.555.8
Gemini 3.1 Pro Preview55.2
Loading Atlas data…