Atlas

Benchmarks

← All benchmarks

VitaBench 2.0 - Rewrite (Agentic Memory)

Agents · 2026-05-26

This VitaBench 2.0 condition evaluates personalized long-term agents on the Chinese Rewrite scenario with the Rewrite memory setting, corresponding to Agentic Memory on the official leaderboard. It reports single-trial Avg@1 by first averaging subtask rewards within each user sequence and then averaging across users.

Top models (higher is better)

ModelScore
Gemini 3.1 Pro Preview50.2
Qwen3.7-Max47.6
GPT-5.547.4
Opus 4.846.3
Macaron-V1-Venti46.0
GLM-5.243.1
MiniMax M339.4
Loading Atlas data…