Hugging Face’s trending models span local assistants, speech transcription, 3D generation and large multimodal systems.

First snapshot, 4 November 2026; these are trending entries, not necessarily new releases.

• LFM3-1.7B: local assistants and tool use; 1.73B parameters; Liquid’s LFM 1.0 license. Its GGUF variant supports local CPUs and GPUs; Ollama lists it.
• Nemotron 3.5 Transcribe: multilingual speech transcription; 30B total, 4B active; OpenMDW 1.1 license. Consumer-hardware suitability not established by the card.
• DeepSeek V4.1 Flash 1017: reasoning, coding and visual tasks; size not stated in this card; MIT license. Consumer-hardware suitability not established.
• Mistral Large 4: multimodal reasoning and coding; 1T total, 52B active; Mistral’s open-weight license restricts model-as-a-service use. Its example uses eight GPUs; Ollama lists it, though local consumer suitability is not established.
• Idol: turns an image into a 3D object; about 1.8B parameters for the full pipeline; Apache 2.0 model, with separately gated encoder components. The card reports 7.2 GB peak GPU memory and tests consumer GPUs.
• Pika H3 Flex: MiniMax-H3 adapters for video with synchronized audio; parameter count not stated; MiniMax H3 Community License. The card reports H100 80GB tests, not consumer-hardware requirements.

MIT and Apache 2.0 are OSI-approved licenses; the other named licenses are distinct open-weight or model licenses and should not be assumed to be OSI open source. Ollama listing alone does not establish that a model runs locally on a particular computer.