NVIDIA, Cohere, and Sarvam Lead Wave of Open Source AI Model Releases

TLDR

NVIDIA released Nemotron-3-Super-120B, a mid-sized model with 120B total parameters featuring a 1M context window and LatentMoE architecture with NVFP4 quantization during pre-training.

Cohere released its Transcribe model under Apache 2.0 license, while Indian startup Sarvam released flagship models demonstrating substantially improved performance for Indic languages.

Key Takeaways

• NVIDIA releases 120B Nemotron with 1M context, Cohere open-sources transcription model under Apache 2
• 0, and Indian startup Sarvam ships 105B Indic model