SenseTime launches SenseNova V6

Background

The release combined two directions: a multimodal model that handles text, images and video together, and reasoning models that work through steps before answering.

What happened

SenseTime unveiled SenseNova V6 on 10 April 2025 ^[1]. It is a native MoE multimodal model with over 600 billion parameters ^[2]. The series included V6 Reasoner and the real-time V6 Omni model ^[2].

Key details

The model supports multimodal chains of thought up to 64K tokens and can analyse 10-minute videos ^[2]. SenseTime described it as China's most advanced multimodal model with the lowest cost in the industry ^[2]. These are the company's own descriptions.

A new SenseChat app based on V6 entered preview ^[2]. SenseTime also announced RMB 100 million in subsidies for embodied intelligence and AIGC developers ^[2].

Impact

The release moved SenseTime's line to a native multimodal reasoning model with a real-time variant. For readers, the event shows how the company tied its models to a consumer chat app and to support for developers in new fields at the same time.

Terms used here

A native multimodal model is trained to handle several kinds of input from the start, which differs from adding a picture reader to a model that was trained only on text. A reasoning model is one that writes out intermediate steps before its final answer. Real-time models handle speech and images while a conversation is happening, rather than waiting for a complete request.

An app built on a model is the product layer that users interact with, while the model is the underlying system. Subsidies for developers are payments or support offered to people building products in a given area, meant to encourage work in that field. A preview release gives a small group of users access before a wide launch, so their feedback can shape the final product.