Hugging Face AI Updates: October 28, 2026
Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled
1. Liquid AI Releases LFM3, Six Open-Weight 2B-Class Models on One Backbone
Hugging Face. Liquid AI announced LFM3 on the Hugging Face blog, a family of open-weight models that branch from a shared LFM3-1.7B-Base: a text model, LFM3-VL-2B (images and video), LFM3-Audio-2B, LFM3-Omni-2B, and two “decision” models (d1-2B and d1-omni-2B) that return calibrated answers. The base uses a 24-layer tri-hybrid mix of 13 gated short convolutions, 5 sliding-window attention layers and 6 global attention layers, was pre-trained on 22T tokens, and supports a 256K context window. Liquid AI says LFM3-1.7B outperforms LFM2.5-2.6B with 35% fewer parameters, scoring 38.42 on WebArena-Verified against 3.02, and fits in about 1 GB at 4-bit precision. New in-house vision and audio encoders are included, and every model can be fine-tuned on a single GPU and runs in Transformers. Source
2. Ai2 Benchmark Finds LLM Novelty Judges Unreliable
Hugging Face. Ai2 published “Old Ideas, Novel Problems” on the Hugging Face blog, a study with a benchmark, data and code for testing LLMs that judge whether research ideas are novel. The benchmark draws on ICLR 2026 peer reviews: 154 high-novelty ideas from top-rated accepted papers, 145 lower-novelty ideas from bottom-rated rejected papers, and 154 LLM-generated ideas. Across six judges (GPT-5.1, GPT-5.2, GPT-5.4, Claude Sonnet 4.5, Claude Opus 4.5 and Claude Opus 4.6), changing a few words in a prompt shifted pairwise accuracy by more than 50 percentage points. Retrieval and extra reasoning did not consistently help, and two purpose-built novelty systems performed worse than a simply prompted Claude Opus 4.6 baseline. Source