Terminal-only code localization
According to the model card, CodeScout performs repository-level code localization using only a standard Unix terminal, without static analysis or repository graphs.
Open Source Model Profile · OpenHands
CodeScout-1.7B
CodeScout-1.7B is a 1.72B-parameter Qwen3 text-generation fine-tune from OpenHands. According to the model card, it is an RL-trained code search agent for terminal-only repository localization.
- Publisher
- OpenHands
- Task
- text-generation
- Model type
- qwen3
- License
- apache-2.0
- Library
- transformers
- Publication status
- Approved for indexing
Model overview
CodeScout-1.7B is published by OpenHands as a Qwen3 text-generation model. The captured configuration identifies Qwen3ForCausalLM and Safetensors metadata reports 1,720,574,976 parameters. According to the model card, it belongs to the CodeScout family of RL-trained code search agents derived from Qwen3-1.7B.
Recorded capabilities
Two-stage RFT plus GSPO training
According to the model card, Qwen3-1.7B is warm-started with rejection fine-tuning and further trained with GSPO reinforcement learning on multi-level F1 rewards.
Reported localization gains
According to the model card, the publisher reports absolute F1 gains over Qwen3-14B and competitive function F1 with larger Qwen3 variants; these are publisher claims, not Ethen measurements.
Use cases in the source record
- Repository-level code localization experiments that navigate files, classes, and functions with Unix commands such as rg, find, grep, and ls.
- RL code-search research using the documented OpenHands-Bash scaffold with Terminal and LocalizationFinish tools, scoped to Python repositories.
Limitations and unknowns
- According to the model card, the model is designed for code localization, not code editing or issue resolution, and was trained and evaluated exclusively on Python repositories.
- According to the model card, performance may vary on repositories significantly different from the training distribution and optimal use requires the OpenHands-Bash scaffold.
- No independent evaluation results were extracted from this record; reported F1 figures are publisher claims.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: OpenHands/CodeScout-1.7B
Captured: Unknown. Processed: 2026-09-07T19:35:50.754627+00:00.
CodeScout-1.7B 📄 Paper • 💻 Code • 🤗 Collection Compact yet powerful — outperforms 8× larger Qwen3-14B using only a Unix terminal. CodeScout-1.7B is part of the CodeScout family of open-source RL-trained code search agents. CodeScout models achieve state-of-the-art repository-level code localization using nothing more than a standard Unix terminal — no static analysis, no repository graphs, no language-specific tooling. Key Highlights Outperforms 8× larger Qwen3-14B with absolute F1 gains of 11–18% for files and 10–15% for functions Competitive with 18× larger Qwen3-32B (Thinking) , surpassing it by 3–6% in function F1 Matches Rep…
F001F002F003F004F005F006F007F010F012F013F014F015F016F017F018F019