Hugging Face expands native open-weight architecture support and introduces an ExportedModel interface for moving compatible checkpoints into supported inference runtimes.
Primary sourceHugging Face — Transformers v5.23.0 official releaseSource published 2026-11-13
Cosmos3-Nano variants generate multi-view driving observations, simulator-conditioned video and robotic-action forecasts; licenses and physical fidelity still require scrutiny.
Primary sourceNVIDIA — Cosmos 3 interactive world simulation technical blogSource published 2026-11-05
A Qwen3.8-27B supervised fine-tune exposes how post-training changes political response framing, but the reported improvements depend on the authors' own rubric and judge.
Primary sourceNous Research — Arete-27B official technical reportSource published 2026-11-06
Baseten releases a Rust/CUDA serving engine for DeepSeek-V4.1-Flash. Its Blackwell throughput is vendor-reported, and Elastic License 2.0 restricts managed hosting.
Primary sourceBaseten Labs — DeepSeek Ultrafast official repository and READMESource published 2026-11-13
Updated quantized weight files target better local model accuracy at similar sizes; Unsloth's 5–10% comparison remains a publisher-reported benchmark claim.
Primary sourceUnsloth — Qwen3.8-27B Dynamic v3.1 announcement on Hugging FaceSource published 2026-11-13
Google outlines local ASR and TTS stacks around Parakeet, PocketTTS, Phonon and Kokoro, with separate Android-system and cross-platform model paths.
Primary sourceGoogle AI Edge Team / Hugging Face — On-Device Speech Recognition and Text-to-Speech in Google AI Edge GallerySource published 2026-11-12
The local-inference stack gives downstream C/C++ integrations an explicit major, minor and patch compatibility policy; it is not a new model checkpoint.
Primary sourceggml-org — llama.cpp v1.0.0: A stronger foundation for the local AI ecosystemSource published 2026-11-13
The openly documented 8,115-question benchmark tests whether AI helpers recognize a mistaken proposed solution and redirect toward the user's actual goal.
Primary sourceAllen Institute for AI — XYBench: Helping with the problem behind the questionSource published 2026-11-13
Researchers add independently trained experts to a frozen Qwen3.5 MoE and combine the skills in one checkpoint, avoiding a standard adapter-merging bottleneck.
Primary sourceTNG Technology Consulting — Bring Your Own Expert! Extending MoE Models with New Domain ExpertiseSource published 2026-11-10
Four public GGUF/NVFP4 repositories target NVIDIA GPUs, but host-memory offload and custom model licenses make the practical deployment story more nuanced.
Primary sourceEdge0 — official Edge0-145B Q3_K model card, weights and licenseSource published 2026-11-13
A new public benchmark evaluates 21 decision and reasoning systems across 11 domains and separately tests option-order stability and missing-answer recognition.
Primary sourceLi et al. — SO-Arena: Benchmarking the Capability Spectrum and Reliability of System One Models, arXivSource published 2026-11-11