Meta has released Muse-Glimmer-30B, a 29.6B dense model with vision capabilities under the Apache 2.0 license.
The release includes GGUF checkpoints on Hugging Face and support via a pull request to llama.cpp.
Meta describes the model as an open agentic solution.
Meta has released Muse Glimmer, a 30-billion parameter dense multimodal model optimized for always-on local agent workflows. The weights are available under the Apache 2.0 license.
Google has open-sourced its WeatherNext 2 and WeatherNext Cyclones AI models to improve cyclone forecasting accuracy. The models achieve state-of-the-art results in predicting track, intensity, and wind structure, providing forecasters with an average of one extra day of predictive accuracy compared to current methods.
Lawrence Berkeley National Laboratory has deployed Meta’s Segment Anything Model 3 (SAM 3) and DINOv3 within the SYNAPS-I initiative to automate X-ray and neutron data analysis. Fine-tuned on scientific imaging data, these models replace manual segmentation bottlenecks with a pipeline that delivers semantically labeled 3D volumes in approximately 15 minutes.
Ohio State University's NLP team has released QUEST-35B, an open-source Deep Research agent trained on approximately 32 H100 GPUs using 8,000 synthetic samples. The team open-sourced the training recipe, code, weights, and datasets, with benchmark results showing competitive performance compared to leading closed-source Deep Research systems.
Researchers at Ohio State University trained QUEST-35B, a Deep Research agent, using approximately 32 H100 GPUs and 8,000 synthetic samples. They open-sourced the training recipe, code, weights, and datasets, with benchmark results showing competitive performance compared to leading closed-source Deep Research systems.
We use cookies to measure traffic and improve the site. You can accept or decline analytics cookies. Privacy policy