Unverified50% confidenceFactExact time
Nemotron 3.5 Super VL replaces a pure attention decoder with Mamba state-space blocks, uses a self-developed RADIO vision encoder, and natively supports NVFP4 four-bit quantization on the Blackwell architecture.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedNemotron 3.5 Bolt's architecture adopts LatentMoE mixing Mamba-2 and Transformer, and the vision side uses the C-RADIO v4-H encoder.75% similarUnverifiedNemotron-H 是 NVIDIA 于 2025 年发布的混合架构系列模型，将 Mamba 状态空间模型层与标准 Transformer 注意力层交替堆叠71% similarUnverifiedNVIDIA's Nemotron-H replaces almost all attention layers with Mamba-2 layers.67% similarUnverifiedThe official illustration for Nemotron 3.5 Super VL shows the model simultaneously connected to four types of tasks—code, charts, credit memo data extraction, and video analysis—with the design goal of serving as the reasoning hub of multimodal agents.62% similarUnverifiedNemotron forms a tiered product matrix, with the local MoE version suited for everyday tasks and privacy scenarios, and the Nemotron Ultra cloud version suited for complex reasoning tasks.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1047815API
curl https://kongchang.com/api/v1/knowledge/claims/1047815MCP
get_claim(id=1047815)