Inference
Meituan LongCat Team

VITA-Bench

An interactive real-world agent benchmark grounded in practical consumer-service tasks such as delivery, in-store consumption, and online travel workflows.

ModelsModels with a result.12
Top resultBest result on this test.47.9%Qwen3.7 Max
Top-3 spreadPoints from first to third.3.6 pts
YearYear of release.2025

Result and price

020406080100$0.3$1$3$10$30RESULTOUTPUT PRICE PER 1M TOKENS · LOG SCALE

Results

12 results
#ModelResultThe score from the source.
01Qwen3.7 Max47.9%
02Qwen3.7 Plus45.6%
03Qwen3.6 Plus44.3%
04Qwen3.5 397B43.7%
05Agents-A1-4B40.3%
06Agents-A138.8%
07Qwen3.6-35B-A3B35.6%
08Claude Opus 4.523.3%
09LongCat-Flash-Lite-Sparse21.7%
10DeepSeek V3.218.5%
11Claude Sonnet 4.517.0%
12GLM-4.715.5%