世界模型有触觉了！50万小时视频，训出首个隐式触觉世界动作模型
团队用50万小时视频让模型学会预测触觉，这在之前没人做到过。它能帮机器人提前感知物体触感，操作更稳。
研究人员利用50万小时视频数据训练出首个隐式触觉世界动作模型。该模型能够根据视觉输入生成对应触觉信号。在机器人操作任务中，它实现了比纯视觉模型更精准的动作预测。这项研究将触觉感知融入了世界模型框架。
This report is categorized under AITOP's "Models" desk. Key entities and technical topics under continuous tracking include 触觉世界模型, 50万小时视频, 世界模型, 多模态感知.
Source attribution: curated from 量子位 (@henry).
1000 个 Claude 智能体自己拆任务、互相校对，11 天干完了人类团队要干几年的形式化证明，还顺手证明了 3 万多个定理。做智能体编排、Lean 或数学 AI 的朋友都该看看这个案例。
Google 把 Gemini 4 Hydrogen 推到了免费用户的自动路由里，桌面、手机、网页全都有。不用付费就能在合适的 prompt 下用到它，想知道路由怎么判断的可以看官方文档。用 Gemini 的朋友今天打开 app 就能碰运气。
Day-zero vLLM support plus single-node GB200 fit means operational cost of serving this model is far lower than typical 594B-scale deployments. Engineering teams running self-hosted inference should benchmark the 630 tok/s TP4 figure against their current MoE serving setups, since the DSpark speculator and NVFP4 experts are the levers behind it. Teams needing 1M-token multimodal context without multi-node clusters have a concrete new option to evaluate.
Thinking Machines 发了个 500B 的多模态 MoE，vLLM 发布当天就能跑，还带了 FP4 量化和投机解码优化。想自部署大模型的人可以看看集成细节，不用等闭源 API。
Mira Murati 的 Thinking Machines 把 inkling 2 small 放上了 Together AI，594B 参数只激活 17B，还有 1M 上下文。写代码、跑长任务、看图都能覆盖，推理成本却按小模型算。做 agent 或长文档处理的开发者可以直接去 Together AI 试。
License terms, not model size, are the real differentiator here: Apache 2.0 lets enterprises ship commercial products without the compliance overhead that Llama's custom license imposes. Teams evaluating self-hosted models for cost control or data privacy can fine-tune and deploy Inkling 2 Small with no royalties and no obligation to release source code. Engineering leads comparing open-weight options should shortlist it alongside other permissively licensed releases rather than treating all "open" models as equivalent.
