Native MoE-based multimodal foundation model with over 600 billion parameters, launched at SenseTime Tech Day. Features multimodal long chain-of-thought up to 64K tokens, global memory, and reinforcement learning advances. V6 Reasoner surpasses GPT-o1 and Gemini 2.0 Flash Thinking on deep reasoning tasks. V6 Omni is China's first model supporting in-depth 10-minute video analysis.

Model Details

Architecture MOE
Parameters 600B+

Variants

Name Parameters Notes
SenseNova V6 Core multimodal foundation model
SenseNova V6 Reasoner Deep reasoning variant surpassing GPT-o1
SenseNova V6 Omni Lightweight full-modal interactive model with video analysis
frontiermultimodalreasoningmoe