Liquid AI has released LFM2.5-24B-A2B, a 24 billion parameter Mixture-of-Experts (MoE) model designed for on-device agents. This model activates approximately 2 billion parameters per token and is optimized for multi-step tool use rather than general chat. It demonstrates strong performance on agentic and tool-use benchmarks, outperforming larger models in specific tasks like AutomationBench, while larger models like Qwen3.6-35B-A3B show better results in knowledge and coding tasks. LFM2.5-24B-A2B is capable of running on consumer hardware such as Apple M5 Max and AMD Ryzen AI Max+ processors, offering fast local tool calling and instruction following. AI
IMPACT Enables more capable AI agents to run efficiently on consumer hardware, potentially accelerating on-device AI applications.
RANK_REASON Frontier-lab model release with system card.
Read on Mastodon — mastodon.social →
- AMD Ryzen AI Max+ 395
- Apple M5 Max
- AutomationBench
- Gemma 4 26B A4B IT
- Hugging Face
- LFM2.5-24B-A2B
- Liquid AI
- mixture of experts
- NVIDIA B300
- Qwen3.6-35B-A3B
- SWE-bench Verified
AI-generated summary · Google Gemini · from 4 sources. How we write summaries →
