Understand this story
Ant inclusionAI open-sourced the DR-Venus-4B-SFT and DR-Venus-4B-RL model checkpoints on April 22, 2026. According to its new GitHub repository, DR-Venus is a 4B-parameter deep research agent trained entirely on open data, built on the Qwen3-4B-Thinking-2507 base, supporting the search + visit tool protocol with up to 200 tool calls, a training context of 200K+ and inference at 256K. The project has open-sourced the full SFT, RL, and Inference pipeline code as well as the SFT/RL and GGUF weights. DR-Venus-4B-RL reaches 37.7 on BrowseComp and 74.7 on BrowseComp-ZH, improving by 2.3 and 2.0 respectively over the SFT version. There are currently no further developments.
AI-generated from public reports
Developments
1 confirmed developments
- Source:English text is being prepared.
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 17
Ant Group's inclusionAI: New GitHub RepositoryFirst-party
Ant Group's inclusionAI Open-Sources DR-Venus: A 4B-Parameter Deep Research Agent Trained on Only 10K Open Data
Ant Group's inclusionAI has open-sourced DR-Venus, a 4B-parameter deep research agent trained entirely on open data. Built on the Qwen3-4B-Thinking-2507 base model, it supports the search + visit tool protocol, allows up to 200 tool-call steps, and uses a training context of 200K+ and inference context of 256K. DR-Venus-4B-RL achieves 37.7 on BrowseComp and 74.7 on BrowseComp-ZH, improving by 2.3 and 2.0 respectively over the SFT version. The project has open-sourced the full SFT, RL, and Inference pipeline code, along with the SFT/RL and GGUF weights.
AI Trends