Mamba

Evidence (7)

• 6Hugging Face Blog·10/6/2026model_release
  Falcon-Emirati: An LLM Tuned for Emirati Arabic

  Hugging Face introduces Falcon-Emirati-7B, a dialect-specialized model built on Falcon-H1-Arabic to better understand and generate Emirati Arabic, including local vocabulary, tone, and cultural context. The article explains the challenges of adapting a general Arabic model to a spoken dialect with limited written data and heavy reliance on idioms, proverbs, and cultural nuance.

• 7InfoQ·8/29/2026research
  FreeToken Improves MoE Inference on Consumer Hardware

  Researchers at UC Berkeley and MIT introduced FreeToken, an open-source inference engine designed to improve the efficiency of Mixture-of-Experts models on consumer hardware. It uses dynamic scheduling and weight management optimizations to speed up decoding and support self-hosted reasoning systems at the edge.

• 8Hacker News·8/19/2026breakthrough
  Ornith-1.5 advances self-improving open models

  Ornith-1.5 is a new family of open models built around an end-to-end self-improvement loop that generates its own tasks, scaffolds, and solution rollouts for reinforcement learning. The release spans 397B MoE, 35B MoE, and 9B dense variants, with reported benchmark results that match or exceed several larger open-source and proprietary models on reasoning, coding, and agentic tasks.

• 8Simon Willison·7/27/2026model_release
  Moonshot releases Kimi K3 open weights, but deployment is limited

  Moonshot AI has released the open weights for Kimi K3 on Hugging Face, making one of its largest language models available for self-hosting. The model targets long-horizon coding and knowledge work, but its 2.8-trillion-parameter MoE design and heavy hardware requirements mean only a small number of organizations can run it themselves.

• 7Hacker News·7/16/2026model_release
  German consortium releases Soofi S, an open 30B model

  A German research consortium led by the KI Bundesverband has released Soofi S 30B-A3B, an open language model trained on Deutsche Telekom's Industrial AI Cloud. The model uses a hybrid Mamba-Transformer mixture-of-experts design and is reported to outperform other fully open models on English, German, and programming benchmarks, while maintaining high throughput on long contexts.

• 8Hugging Face Blog·4/28/2026model_release
  NVIDIA Launches Nemotron 3 Nano Omni for Long-Context Multimodal AI

  NVIDIA introduced Nemotron 3 Nano Omni, an omni-modal model for document analysis, speech recognition, long audio-video understanding, and agentic computer use. The model claims strong benchmark results on document, video, audio, and voice tasks, along with higher throughput and efficiency than comparable open multimodal models.

• 8Hugging Face Blog·1/5/2026model_release
  Hugging Face Debuts Falcon-H1-Arabic

  Hugging Face announced Falcon-H1-Arabic, a new Arabic language model family built on a hybrid Mamba-Transformer architecture. The release includes 3B, 7B, and 34B models with expanded context windows up to 256K tokens and aims to improve long-context understanding, dialect handling, and reasoning in Arabic NLP.