SuperAGI/SuperScout-7B

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 15, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

SuperAGI/SuperScout-7B is a 7.6 billion parameter searcher model, fine-tuned from Qwen2.5-Coder-7B-Instruct, designed for repository-level software issue localization. It explores codebases, identifies implicated files, and attempts to generate failing reproductions, emitting a structured handoff document. This model excels at acting as the front-end for automated software repair systems, localizing issues across Python, Go, TypeScript, and JavaScript, and demonstrating language transfer capabilities to other programming languages.

Loading preview...

SuperScout-7B: A Specialized Software Issue Searcher

SuperScout-7B is a 7.6 billion parameter model developed by SuperAGI, specifically engineered to act as a "searcher" for repository-level software issues. It is a LoRA fine-tune of Qwen2.5-Coder-7B-Instruct, trained on nearly 20,000 search-phase demonstrations.

Key Capabilities & Features

  • Issue Localization: Given a software issue and a repository, it explores the codebase to localize implicated files.
  • Reproduction Generation: Attempts to write a failing reproduction for the identified issue.
  • Structured Handoff: Emits a structured document containing identified files, reproduction steps, and notes, designed for downstream "fixer" models.
  • Multilingual Support: Trained on Python, Go, TypeScript, and JavaScript, it also demonstrates strong language transfer capabilities to other languages like those in the SWE-bench Multilingual evaluation.
  • System Integration: Forms the front-end of the SuperScout system, which verifies reproduction claims and routes tasks to various fixer models (e.g., GPT-5.2, Claude Opus 4.6).
  • Optimized Decoding: Achieves significantly better performance (2.65x gain in file localization) with sampled decoding at temperature 0.9 compared to greedy decoding.

When to Use SuperScout-7B

  • Automated Software Repair: Ideal for systems requiring an initial phase of issue diagnosis and localization before a fix is generated.
  • Cost-Effective Issue Scouting: The SuperScout system, leveraging this model, matches top-tier fixer models in solve rate at a fraction of the cost.
  • Repository Exploration: For tasks that involve understanding and navigating large codebases to pinpoint problem areas.

Limitations: SuperScout-7B is a searcher only; it does not generate code fixes. Its reproduction claims require verification by a sandbox system, as a significant portion may be false positives.

Finetunes
1 models