Model Releases

DR-Venus: An edge-level deep research agent based on open data

Original title:inclusionAI/DR-Venus

AI summary

DR-Venus is a 4-billion-parameter deep research agent trained with only 10,000 open data samples, based on the Qwen3-4B-Thinking-2507 architecture, supporting 200-step tool calls and over 200,000 tokens of context. Through two-stage training with supervised fine-tuning and reinforcement learning, it sets a new performance benchmark for small models on multiple deep research benchmarks including BrowseComp and GAIA. Its SFT version has surpassed most comparable open-source models, while the RL version further improves long-horizon task reliability and tool-use calibration by 2-3 percentage points. The project has fully open-sourced the model, code, and training pipeline.

Key Takeaways

  • DR-Venus is a 4-billion-parameter deep research agent trained with only 10,000 open data…
  • Through two-stage training with supervised fine-tuning and reinforcement learning,it sets a new…
  • Its SFT version has surpassed most comparable open-source models,while the RL version further…
  • The project has fully open-sourced the model,code,and training pipeline.

Quick Facts

Published
Updated
Source
蚂蚁 inclusionAI:GitHub 新仓库
Category
Model Releases
AI score
69
Read the original Read the full article at the original source
Why it matters

4B 参数、仅用 1 万条公开数据就能在多个 deep research benchmark 上碾压 8B 对手,蚂蚁 inclusionAI 这次证明了小模型做 Agent 的关键不在参数量而在数据管线,做端侧 Agent 的团队值得拆一下它的 SFT+RL 流程。

Source & copyright

This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.

Original article · AIHOT entry

How to Cite

When citing this article, keep the title, date, and permalink, and credit Orange News as the source.

Orange News: DR-Venus: An edge-level deep research agent based on open data. April 20, 2026. https://new.nat6.net/en/news/cmorb7ik2006islhfoemox8w3/

View all →