acqrn/FastContext-1.0-4B-SFT

TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jul 1, 2026License:mitArchitecture:Transformer Open Weights Cold

FastContext-1.0-4B-SFT by acqrn is a 4 billion parameter supervised fine-tuned subagent model, built on the Qwen3-4B-Instruct backbone, designed for efficient repository exploration within LLM coding agents. It specializes in issuing parallel read-only tool calls (READ, GLOB, GREP) to return compact file paths and line ranges, significantly reducing token consumption and improving resolution rates for main coding agents. This model acts as a dedicated explorer, offloading the search burden from primary LLMs and providing focused context for coding tasks.

Loading preview...

FastContext-1.0-4B-SFT: An Efficient Repository Explorer for Coding Agents

FastContext-1.0-4B-SFT is a 4 billion parameter subagent model, part of the FastContext family, specifically designed to optimize repository exploration for larger LLM coding agents. Developed by acqrn and built on the Qwen3-4B-Instruct backbone, this model separates the task of code exploration from problem-solving, allowing main agents to focus on task execution with clean, grounded evidence.

Key Capabilities & Features

  • Dedicated Repository Exploration: Acts as a specialized subagent, invoked by a main coding agent to handle code search and discovery.
  • Parallel Tool Calling: Efficiently issues multiple READ, GLOB, and GREP tool calls in parallel to quickly locate relevant code.
  • Context Optimization: Returns compact file paths and line ranges, significantly reducing the main agent's token consumption (up to 60% savings) and providing focused context.
  • Improved End-to-End Performance: Integrates with main agents like GPT-5.4, GLM-5.1, and Kimi-K2.6 to improve resolution rates on benchmarks like SWE-bench Multilingual and SWE-bench Pro by up to 5.5%.
  • Training Methodology: Bootstrapped from strong reference-model trajectories via supervised fine-tuning (SFT) and refined with task-grounded reinforcement learning (RL).
  • High Context Length: Supports a context length of up to 262K tokens for comprehensive repository analysis.

When to Use This Model

FastContext-1.0-4B-SFT is ideal for developers building or enhancing LLM-powered coding agents where efficient and precise code repository exploration is critical. It's particularly beneficial for scenarios where reducing token usage and improving the accuracy of code-related tasks are primary concerns, allowing the main agent to operate more effectively by offloading the exploratory burden.

Finetunes
1 models