Samsung presented a 16 GB LPDDR5X-PIM memory package at Hot Chips 2026 that delivers 614 GB/s of internal bandwidth to its own compute units, an eightfold increase over the 76.8 GB/s available through…
Liquid AI released LFM2.5-VL-3B, a 3.1B-parameter vision-language model for on-device deployment, averaging 69.4 across 28 vision benchmarks, matching InternVL-3.5-4B and 0.7 points behind Qwen3.5-4B.…
Meta released Muse Glimmer, a 30-billion-parameter open-weight model for local agentic work, with a 4-bit version fitting under 20 GB and tested within 24 GB or 32 GB memory envelopes on consumer hard…
Meta released Muse Glimmer, a 30-billion-parameter multimodal agentic model distilled from Muse Spark, under the Apache 2.0 license, designed to run on a single consumer GPU or Mac with no network cal…
Simon Willison successfully ran MiniMax-H3, a text-to-video model, on an Apple M5 Max MacBook Pro using the PipeNetwork/minimax-h3-mlx Python package, which ports the model to MLX. The process downloa…
Liquid AI released LFM2.5-2.6B, a 2.6-billion-parameter on-device agentic model that outperforms models up to 4x larger on tool use and instruction following, achieving 220 tokens per second on an App…
The mlx-community/Laguna-S-2.1-oQ2e quantized model running in-process through mlx-vlm on a 128 GB Apple M5 Max achieved a perfect overall score of 1.000 across six tasks, with 40.85 generation tok/s …
A developer known as JustVugg built Colibri, a pure C inference engine that runs the 744-billion-parameter GLM-5.2 model on a laptop with 25 GB of RAM by streaming experts from disk instead of loading…
A proof-of-mechanism study of ontology-amplified distillation found that a Qwen3.6-27B student model, fine-tuned on 47 synthetic English-language preference pairs using Foundation AgenticOS ontology, …