Chinese AI company DeepSeek is about to release its fourth generation big model V4, introducing revolutionary Engram memory architecture and mHC (manifold constrained hyperconnectivity) technology. The new model adopts a sparse MoE architecture, supports 1 million token context windows, reduces memory usage by 40%, improves inference speed by 1.8 times, and natively supports multimodal generation of text, images, and videos.

**Together AI**, as a diamond sponsor of **NVIDIA GTC 2026**, is showcasing its latest research and product innovations at Booth #1213 in San Jose from March 16 to 19. Today’s updates focus on open-source LLMs, voice AI capabilities, production-grade inference, and AI factory infrastructure. Key announcements include **FlashAttention-4** (up to 1.3× faster than cuDNN on NVIDIA Blackwell), the open-source **ThunderAgent** for agentic workloads (delivering a 3.6× throughput improvement), the **ATLAS-2** adaptive learning speculator, and a full-featured voice AI stack supporting real-time speech-to-text and text-to-speech. Together AI demonstrates how enterprises can transition from AI experiments to production deployment in minutes using its GPU clusters and inference platform.