CriticalCapability
Open inMute releases CoderForge-Preview: SOTA open dataset for training coding agents
Action Required
Researchers can now leverage a large, high-quality dataset to accelerate the development of more efficient and capable coding agents.
AI Impact Summary
Open inMute is releasing CoderForge-Preview, a massive open dataset of 258,134 coding agent trajectories generated using Qwen3-Coder-480B. This dataset, with a median length of 41,000 tokens and a total of 6.7 billion tokens, is designed to accelerate research in efficient coding agents. The dataset's focus on successful trajectories and long-context coverage makes it a valuable resource for training and evaluating coding agents, particularly for models like Qwen-3.
Open inMute releases CoderForge-Preview: SOTA open dataset for training coding agents
Affected Systems
Qwen3-Coder-480B
- Date
- Date not specified
- Change type
- capability
- Severity
- critical