Liquid AI Releases LFM2.5-2.6B: An On-Device Agentic Model With 128K Context, Tool Calling, And Open Weights
2.6B parameters, 128K context, tool calling—all on a MacBook. Liquid AI's LFM2.5 brings agentic capability to the edge.
Why it matters
Liquid AI releases an open-weight agentic model small enough to run locally with full multi-step reasoning and tool-calling capability. This shifts the frontier conversation from cloud-dependent agents to edge deployment, and the 220 tok/s decode speed on consumer hardware makes on-device agent workflows practically viable.
The key facts
8 to knowModel: LFM2.5-2.6B (2.69B parameters)
Architecture: 22 double-gated short convolution blocks + 8 GQA blocks across 30 layers
Context window: 131,072 tokens (128K)
Performance: 220 tokens/s on M5 Max, under 2.5 GB memory
Capability: agentic (planning, tool calling, multi-step task completion)
Weights: open, released in GGUF, MLX, and ONNX formats
Inference location: entirely on-device
Publisher: Liquid AI
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: Liquid AI released LFM2.5-2.6B, an agentic model that plans, calls tools, and completes multi-step tasks entirely on-device. The 2.69B parameter model pairs 22 double-gated short convolution blocks with 8 GQA blocks across 30 layers, handles 131,072 tokens of context, and decodes at 220 tokens/s on…