🎙 EP 378: Meta Releases Muse Spark Open & NVIDIA Debuts Nemotron 3.5 Super VL

Podcast: AI Fire Daily · AIFire.co

Opublikowano: 2026-10-16

Meta launched Muse Spark Open, its largest U.S.-developed open-weight model featuring an 816B sparse Mixture-of-Experts architecture (42B active parameters), a 1M-token context window, and native Apple M5 Ultra optimization. Meanwhile, NVIDIA released Nemotron 3.5 Super VL, a 120B parameter MoE model that activates 12B parameters per token to deliver 5× faster multimodal reasoning and throughput for enterprise agent workloads. We’ll talk about: High-capacity sparse MoE model optimized for Apple M5 Ultra Mac Studios, offering 1M-token context and leading open benchmark performance.Multimodal open model achieving 600+ output tokens per second and 5× higher inference throughput for...