AITOPAI热报
Models

Om AI发布端侧原生VLX模型：3B小参数实现物理世界精准感知

3B模型碾压英伟达谷歌后，Om AI端侧原生VLX模型：小参数实现物理世界精准感知

量子位@田, 晏林
Blog/Media
Read the original
Why we picked it

Om AI的3B端侧模型VLX，能在设备上精准感知物理世界，性能还超过英伟达和谷歌。

Om AI发布了端侧原生VLX模型，参数量仅3B。官方称该模型在物理世界感知任务上超过英伟达和谷歌的方案。VLX面向设备端部署，用3B小参数实现精准感知。

Image · 量子位
Original · 量子位

3B模型碾压英伟达谷歌后，Om AI端侧原生VLX模型：小参数实现物理世界精准感知

从云计算到云原生，产业真正的跃迁从来不是简单堆算力，而是找到更适合新世界的架构。

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Models" desk. Key entities and technical topics under continuous tracking include VLX, Om AI, multimodal, on-device deployment.

Source attribution: curated from 量子位 (@田, 晏林).

#VLX#Om AI#multimodal#on-device deployment

Keep reading

Recent stories with the same tag or section
ModelsBuildersSelected01:02
Liquid AI open-sources LiquidCodec for low-latency on-device voice AI

Codec throughput, not model size, is often the hidden bottleneck in streaming voice agents, and parallel feature generation attacks exactly that constraint. Builders of on-device speech assistants and edge inference tools should benchmark it against codecs like EnCodec or SoundStream for latency and quality. The M4 Max figures also hint it fits Apple Silicon laptops and phones well, useful for teams shipping local-first voice features.

#Liquid AI#LiquidCodec#speech synthesis#open-source models
Gorden Sun@Gorden_SunSource
ModelsBuildersMulti-source01:01
Chinese contentGemini 4 Hydrogen 实测：能力升级的 Gemini 4 Flash，价格更普惠

一位博主实测发现 Gemini 4 Hydrogen 其实就是升级版 Flash，代码和长任务能力上去了，在他的视频工作流里直接替掉 Opus 5.5，成本省一半。Gemini Pro 订阅就能用，做视频、跑长任务的可以试试。

#Gemini 4 Hydrogen#Gemini#Opus 5.5#coding assistants
Event
Gorden Sun@Gorden_SunSource
ModelsBuildersMulti-sourceSelected01:01
TurboPairFormer GPU kernels speed up OpenFold3 triangle attention

Bitwise-reproducible gradients combined with a speedup is rare in scientific GPU kernels, since most fast attention implementations sacrifice determinism. Labs running AlphaFold-style training on H100 or H200 clusters can cut per-step time directly by swapping the backend. The comparison against OpenFold3's Triton backend gives a concrete baseline for teams evaluating whether to switch.

#TurboPairFormer#OpenFold3#Lambda#H100
Event
Lambda@LambdaAPISource
ModelsUS vs ChinaMulti-source86°01:01
Chinese contentThinking Machines 开源 Inkling 2 Small，594B 参数对标 DeepSeek

Mira Murati 的 Thinking Machines 把 Inkling 2 Small 开源了，594B 参数但只激活 17B，跑分逼近 DeepSeek-V4.1-Flash-1017。做本地部署或自建服务的开发者可以下载试试，美国开源模型终于有能和国产线掰手腕的了。

#Inkling 2 Small#Thinking Machines#DeepSeek#open-source models
Event
Gorden Sun@Gorden_SunSource