breatic Arena

真实的创作任务,匿名的回答,投完才揭晓是谁。
Real creative tasks, anonymous answers — the models are revealed after you vote.

2026-10-22
Agent 主模型盲评 · Agent model blind vote
DeepSeek V4.1 Flash 1017(思考 high / max)、Claude Sonnet 5.5、Claude Opus 5.5 驱动同一个创作 Agent,先比方案,再比成品。
The same creative agent driven by four models: judge the plan first, then the pictures.