OPEN LINK ↗
// 1h agoMODEL RELEASE
Liquid AI Drops LFM2.5 for Local Agents
Liquid AI released LFM2.5-24B-A2B, a 24B-parameter reasoning MoE with roughly 2B active parameters, designed for fast, private agent workloads on laptops. It is open-weight, supports major local inference stacks, and reportedly outperforms larger models on several agentic benchmarks.
// ANALYSIS
This is a strong bet that useful agents need fast, local tool loops more than maximum parameter counts.
- –Competes with Qwen3.6-35B-A3B on five of eight agentic and tool-use benchmarks
- –Reaches 184 tokens per second on an Apple M5 Max, making continuous local agents more practical
- –Supports llama.cpp, MLX, vLLM, SGLang, and ONNX, reducing deployment friction
- –Its 131K context window and explicit reasoning improve suitability for multi-step workflows
- –Benchmark claims are promising, but early community reports suggest real-world tool calling still needs validation
// TAGS
lfm2.5-24b-a2bllmopen-weightsmoeagenttool-uselocal-first
DISCOVERED
1h ago
2026-10-14
PUBLISHED
1h ago
2026-10-14
RELEVANCE
10/ 10
AUTHOR
0x0SojalSec