PhysShell/FastContext-1.0-4B-SFT

TEXT GENERATIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 2, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

PhysShell/FastContext-1.0-4B-SFT is a 4 billion parameter supervised fine-tuned (SFT) subagent model, based 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 as focused context. This model significantly reduces the main agent's token consumption and improves end-to-end resolution rates by offloading repository search tasks.

Loading preview...

FastContext-1.0-4B-SFT: An Efficient Repository Explorer Subagent

FastContext-1.0-4B-SFT is a 4 billion parameter model, built on the Qwen3-4B-Instruct backbone, specifically engineered as a dedicated subagent for large language model (LLM) coding agents. Its primary function is to efficiently explore code repositories, addressing a major bottleneck in current coding agent workflows where repository exploration consumes a significant portion of token budgets and pollutes the main agent's context.

Key Capabilities

  • Dedicated Repository Exploration: Acts as an on-demand subagent, allowing the main coding agent to focus on problem-solving rather than repository navigation.
  • Parallel Tool Calling: Executes multiple READ, GLOB, and GREP tool calls concurrently to quickly identify relevant code snippets.
  • Context Optimization: Returns compact file paths and line ranges, providing clean, grounded evidence to the main agent and reducing its token consumption by up to 60.3%.
  • Improved End-to-End Performance: Integrates with main agents like GPT-5.4, GLM-5.1, and Kimi-K2.6, improving end-to-end resolution rates by up to 5.5% on benchmarks like Mini-SWE-Agent, SWE-bench Multilingual, SWE-bench Pro, and SWE-QA.
  • Training Methodology: Bootstrapped via supervised fine-tuning (SFT) on reference-model trajectories, covering broad first-turn search, multi-turn evidence gathering, and precise citation generation.

When to Use This Model

FastContext-1.0-4B-SFT is ideal for developers building or enhancing LLM-powered coding agents that require efficient and focused repository interaction. It's particularly beneficial for scenarios where reducing token usage and improving the accuracy of code exploration are critical, allowing the main agent to operate with a cleaner, more relevant context.

Finetunes
1 models