Imagine having access to a comprehensive, fully open-source deep research project that includes not just the agent but also the model, dataset, and training/RL infrastructure. This is exactly what MiroMind has achieved with the release of MiroMind Open Deep Research (ODR) v0.1, a game-changing innovation in the field of artificial intelligence.
ODR v0.1 is the first full-stack, fully open-source deep research project that includes four layers: Agent (MiroFlow), Model (MiroThinker), Data (MiroVerse), and Training/RL (MiroTrain / MiroRL). This means that researchers and developers can now access and build upon the entire stack, from the agent framework to the model series, dataset, and training infrastructure.
The implications of this release are significant. With ODR v0.1, the research community can now focus on advancing the state-of-the-art in deep research, rather than reinventing the wheel. The full-stack openness of ODR v0.1 also ensures reproducibility, making it easier for researchers to validate and build upon existing work.
The numbers speak for themselves. MiroFlow, the agent framework, reproduces 82.4 on GAIA validation, while MiroThinker, the model series, reaches 60.2% on GAIA-Text-103. MiroVerse, the dataset, includes over 147,000 full rollout trajectories, making it an invaluable resource for researchers working on tool-use and web-browsing agents.
If you’re interested in learning more about ODR v0.1, I recommend checking out the overview blog post, which includes tables and results. You can also explore the individual components of ODR v0.1, including MiroFlow, MiroThinker, MiroVerse, and MiroTrain / MiroRL.