LakoMoor/QClaw-4B

VISIONConcurrency Cost:1Model Size:4.5BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Apr 24, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

LakoMoor/QClaw-4B is a 4.5 billion parameter decoder-only transformer model fine-tuned for agentic tasks and tool use. Designed for compatibility with OpenClaw frameworks, it achieves state-of-the-art results in the 4B class on the ClawBench agent benchmark. This model excels at multi-step reasoning and tool calling, making it suitable for compact, efficient agentic pipelines.

Loading preview...

QClaw-4B: A Compact Agentic Model

QClaw-4B, developed by LakoMoor, is a 4.5 billion parameter language model specifically fine-tuned for agentic tasks and tool use. It is designed to integrate seamlessly with OpenClaw-compatible agent frameworks.

Key Capabilities & Performance

Despite its relatively small size, QClaw-4B demonstrates impressive performance, matching or exceeding larger models like Kimi K2.5 and GLM-4.5 on the ClawBench agent benchmark. It achieved an Overall Score of 84.8/100 with a 73.5% pass rate across 1110 tasks, indicating strong capabilities in multi-step planning and tool invocation. The model is considered state-of-the-art in the \u22644B parameter class for agentic workflows.

Training & Architecture

QClaw-4B is built on a decoder-only transformer architecture. Its training involved a curated mixture of:

  • Agentic task trajectories (tool calling, multi-step planning)
  • Instruction-following data
  • Code and structured reasoning

Intended Use Cases

This model is particularly well-suited for:

  • Agentic pipelines utilizing OpenClaw or similar frameworks.
  • Developing tool-augmented assistants that require efficient and compact inference.
  • Research into the capabilities of small models for agentic applications.

It is important to note that QClaw-4B is not intended for safety-critical systems without additional alignment work.