Open Weights
-
New Models
NVIDIA releases Nemotron 3.5 Super VL, an open 120B multimodal model for always-on agents
NVIDIA on October 15, 2026 released open Nemotron 3.5 Super VL, a 120-billion-parameter multimodal model for specialised always-on agents that organisations can post-train and host themselves, with Bolt edge variants lis…
-
Open Source
Meta releases Muse Spark Open weights, an 816B-parameter multimodal MoE built for coding agents
Meta on October 15, 2026 published open weights for Muse Spark Open, an 816B-total / ~42B-active multimodal mixture-of-experts model with a 1M-token context window, also demonstrated on two Apple M5 Ultra Mac Studios via…
-
New Models
Tether releases QVAC TranscribePsy, a 254M-parameter open-weight speech recognizer built to run offline on phones
Tether AI Research has published QVAC TranscribePsy under Apache 2.0, a compact multilingual speech-to-text model distilled from NVIDIA's Parakeet that the company says beats similar-size rivals on accuracy and phone spe…
-
New Models
Liquid AI releases LFM2.5-24B-A2B, an open-weight reasoning model built for fast on-device agents
Liquid AI has published LFM2.5-24B-A2B on Hugging Face, a 24-billion-parameter mixture-of-experts reasoning model that activates about 2 billion parameters per token and is trained to call tools inside agent harnesses. I…
-
Business
Nvidia in early talks to deepen its stake in Reflection AI or buy the open-weight model startup
The Financial Times and Bloomberg report that Nvidia is discussing a larger investment, a computing deal or an acquisition of Reflection AI, days after the startup unveiled its Beam open-weight model. No terms have been …
-
New Models
Mistral announces Large 4 “Le Chonk” ahead of October 27 open-weight release
Mistral has announced Mistral Large 4, nicknamed “Le Chonk,” with full public weights scheduled for October 27 after a limited early-access period. Independent reports describe a roughly 1-trillion-parameter mixture-of-e…
-
New Models
Cohere opens North Small Translate’s weights, but commercial use still needs a license
Cohere releases North Small Translate with downloadable weights, 50-language support and MoE serving options, but public weights remain non-commercial.
-
New Models
DeepSeek V4.1 Flash ships with native multimodality and a new model architecture
DeepSeek releases V4.1 Flash with native multimodality, open weights and new API routing that changes how older Flash and V4 Pro identifiers resolve.