模型多源确认

Odyssey 发布世界模型 Odyssey-3,Pro 版 Physics-IQ Verified 得分 66.1

精选理由

Odyssey-3 能生成会对你操作实时反应的互动世界,Flash 版免费可玩,还顺手在 Physics-IQ 拿了 66.1 的高分,去看看 demo。

Odyssey 发布世界模型 Odyssey-3,其 Pro 版在 Physics-IQ Verified 视频到视频基准上取得 66.1 分,为目前公开报告的最高成绩。模型采用自回归扩散 transformer 架构,根据先前帧和输入动作预测下一帧,使场景能响应每个操作。蒸馏版 Odyssey-3 Flash 可从文本提示生成实时可探索、随移动和事件变化的世界,今天起免费开放体验。用冻结的 Odyssey-3 训练的策略在 20 小时印度道路数据上驾驶了真实车辆,机械臂也能从演示中未出现的抓取失误中恢复。在图生视频任务上,Odyssey-3 Pro 与 FLUX 3 [large] 大致持平。

原文 · rohanpaul_ai

Odyssey released Odyssey-3, a world model whose Pro version posts the highest reported Physics-IQ Verified video-to-video score, 66.1.

It's an autoregressive diffusion transformer that predicts next frames from earlier ones plus incoming actions, letting prompted scenes react to each move.

> Odyssey-3 Flash, a distilled few-step version, turns a text prompt into an explorable world that reacts to movement and events in real time, and anyone can try it free today.

> With Odyssey-3 frozen, a policy trained on 20 hours of Indian road data drove a real car, and a robot arm recovered from missed grasps its demos never showed, though neither result reports success rates.

> on image-to-video Odyssey-3 Pro roughly ties FLUX 3 [large]