A September 14, 2026 arXiv paper demonstrates "capability laundering," in which a weaker unaligned model splits a harmful task into benign subproblems, consults a stronger aligned model on each indepe…
TokenRhythm released NeoHorse-1-9B, a 9-billion-parameter model post-trained from Qwen3.5-9B that scores a 69.04 macro average across ten benchmarks, a 3.44-point gain over the 65.60 Qwen3.5-9B baseli…
A €803 machine built from parts up to nine years old achieves roughly half the generation speed of a €2,392 current machine, and on some models three quarters, according to a measurement by the author…
Ornith AI released Ornith-1.5-35B-A3B, a 35B-parameter mixture-of-experts model that activates about 3B parameters per token, trained via a self-improvement loop that jointly optimizes task generation…
Meta Inc. released Muse-Glimmer-30B, a 30-billion-parameter image-text-to-text agent model that fits in approximately 17GB, along with assistant, GGUF, and ExecuTorch PTE variants on Hugging Face. The…