NVIDIA Releases Nemotron 3.5 Transcribe: 30B MoE, 40-Language ASR
NVIDIAAI · x · 2026-10-29
NVIDIA AI announced that Nemotron 3.5 Transcribe is served on SGLang-Omni from day 0.
- 30B MoE architecture with 4B active parameters
- Transcription across 40 languages
- Served via the standard /v1/audio/transcriptions endpoint, with streaming and custom prompts
- Available on Hugging Face for immediate deployment
Related event: NVIDIA Open-Sources Nemotron 3.5 Transcribe 30B Speech Model(10 posts)→
More from Infra
- lium rents NVIDIA B300s at $4/GPU-hour, H200 prices for Blackwell Ultra — bittingthembits · 2026-10-29
- Microsoft Expected to Generate $8B+ Cash Flow in Q3 While Meta, Google, Amazon Burn AI Capex — Beth_Kindig · 2026-10-29
- Bloomberg: AI infra bond yields hit double digits as credit market nears its limits — asusarla · 2026-10-29
- PsiQuantum unveils low-loss 8×8 optical circuit switch chip for AI datacenters — jwt0625 · 2026-10-29
- For local inference, an RTX A6000 eGPU box beats a $22k+ setup — beffjezos · 2026-10-29
- Boat launches CI runners on cloud VMs claiming 20x scaling on the same budget — zack_overflow · 2026-10-29