AI BriefingLiquidAI releases LFM2.5-24B-A2B, a 24B-parameter reasoning model with 2B active parameters for on-device agents
Key point
The model achieves 184 tokens per second on an Apple M5 Max and scores 48.09 on LiveClaw, outperforming Qwen3.6-35B-A3B.
Details
LiquidAI has released LFM2.5-24B-A2B, the largest and final model in the LFM2.5 family, designed for fast, private agents on-device. It is a general-purpose text-only reasoning model with 24.0B total parameters but only 2.45B active parameters per token, enabling efficient inference.
Performance and Benchmarks
The model is positioned as a best-in-class agent, scoring 48.09 on LiveClaw compared to 38.19 for Qwen3.6-35B-A3B, a model with approximately 50% more total and active parameters. In terms of speed, LFM2.5-24B-A2B runs up to 1.7× faster than comparable MoE models on-device. Specific benchmarks include:
- 184 tokens per second on an Apple M5 Max.
- Up to 21K output tokens per second on a single NVIDIA B300 GPU.
Technical Specifications
The architecture features 40 layers (30 double-gated short convolution blocks and 10 GQA) and 64 experts per MoE block with top-4 routing. Key configuration details include:
- Training budget: 36 trillion tokens.
- Vocabulary size: 128,000.
- Context length: 131,072 tokens.
- Supported languages: English, Arabic, Chinese, French, German, Italian, Japanese, Korean, Portuguese, Spanish, Vietnamese, Thai, Indonesian, Hindi, Russian, and Polish.
The model is available in GGUF format on Hugging Face, with a free demo hosted on LiquidAI's playground.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.