LFM2.5-2.6B

Evidence (6)

• 7Hugging Face Blog·8/20/2026model_release
  LFM2.5 DSpark Draft Models Deliver Faster Inference

  Hugging Face and Liquid AI released DSpark draft checkpoints for three LFM2.5 models, adding speculative decoding to improve inference speed without changing output quality. The company reports up to 3.18x higher GPU throughput, up to 2.87x on-device speedups, and a 57% average reduction in function-calling latency for LFM2.5-2.6B, with day-one support in llama.cpp and SGLang.

• 6Hugging Face Blog·8/19/2026model_release
  Liquid AI releases QAD Q4_0 checkpoints for LFM2.5

  Liquid AI released updated Q4_0 GGUF checkpoints for four LFM2.5 models using quantization-aware distillation. The company says the checkpoints preserve Q4_0 memory and speed while recovering most of the BF16 accuracy lost in quantization, and they are available for use with llama.cpp and other GGUF-compatible runtimes.

• 8Hugging Face Blog·8/4/2026model_release
  Hugging Face introduces LFM2.5-2.6B for on-device agents

  Hugging Face announced LFM2.5-2.6B, a 2.6B-parameter model designed to run capable agents entirely on-device with tool calling and multi-step workflows. The model is positioned for laptops and phones, emphasizing privacy, low memory use, and strong benchmark performance relative to much larger models.

• 7Hugging Face Blog·7/28/2026model_release
  LiquidAI releases LFM2.5 encoders for fast long-context CPU inference

  Hugging Face announced two new encoder models, LFM2.5-Encoder-230M and LFM2.5-Encoder-350M, designed for long-context inference with strong accuracy and low latency on CPU. The models support up to 8,192 tokens, are trained with masked language modeling, and are positioned for use cases such as classification, intent routing, safety filtering, and PII detection.

• 5Hacker News·7/20/2026product_launch
  Nativ lets Mac users run open models locally

  Nativ is a curated library for selecting and running open AI models locally on Mac hardware. It recommends partner models based on the user's machine and highlights options from Google, Cohere, and Liquid AI with different sizes and context windows.

• 5Hugging Face Blog·2/20/2026product_launch
  Train small AI models for free with Unsloth and Hugging Face Jobs

  Hugging Face describes a workflow for fine-tuning the small LFM2.5-1.2B-Instruct model using Unsloth on Hugging Face Jobs, with free credits offered to eligible users. The post highlights faster training, lower VRAM usage, and agent-based setup through tools like Claude Code and Codex.