LFM2.5 is a family of hybrid models from Liquid AI designed for on-device deployment. It builds on the LFM2 architecture with extended pre-training and reinforcement learning.

LFM2.5-24B-A2B

The largest and final model in the LFM2.5 family: a 24B reasoning model that activates only about 2B parameters per token, built for fast, private agents on your own hardware.

ollama run lfm2.5:24b

• Built for agents: Trained with reinforcement learning inside popular agent harnesses. Liquid AI reports a LiveClaw score of 48.09, ahead of Qwen3.6-35B-A3B (38.19), a model with about 50% more parameters.
• Always reasons: Thinks before every answer, and with only 2B active parameters each reasoning token is cheap.
• Long context: 128K tokens.
• Multilingual: English, Arabic, Chinese, French, German, Italian, Japanese, Korean, Portuguese, Spanish, Vietnamese, Thai, Indonesian, Hindi, Russian and Polish.

Use it for agents that reason and call tools. For knowledge-heavy questions, pair it with web search or retrieval.

Use it with your agent

Liquid AI trained LFM2.5-24B-A2B inside OpenClaw, Hermes Agent, Pi and OpenCode. Start any of them with Ollama:

ollama launch opencode --model lfm2.5:24b

ollama launch hermes --model lfm2.5:24b

LFM2.5-8B-A1B

An edge model built for fast, reliable tool calling on consumer hardware.

ollama run lfm2.5:8b

• On-device personal assistant: Designed to power real-life applications, chaining tool calls, and following complex instructions on all devices.
• Compressed performance: Competitive with much larger dense and MoE models on instruction following and agentic tasks.
• Unmatched throughput: Fastest in its size class on both CPU and GPU inference