MIRA
A multimodal deep research agent for controllable image-text report generation. It links four stages—evidence collection, report submission, feedback revision, and final delivery—through a generation contract, supporting traceable integration of textual and visual evidence.
Timeline (last 90 days)
MIRA is a multimodal text-and-image deep research agent aimed at controllable reports.
The core component of MIRA is MIRA-Harness, which translates user requirements into a generation contract and connects four stages: evidence collection, report submission, feedback-driven revision and final delivery.
In the evidence collection stage, MIRA's agent performs multi-step web searches, captures both textual and image evidence, and records their sources, laying the foundation for subsequent traceability.
MIRA post-trains a foundation model on MIRA-Data, enabling the model to understand the constraints of the generation contract and learn to actively use image evidence during search.
MIRA achieves state-of-the-art performance among open source agents on 7 representative multimodal search and deep research agent benchmarks.
The text-and-image reports generated by MIRA align with user requirements, and both textual and visual evidence are traceable, allowing users to trace back the original materials underlying each conclusion, including image sources.
A generation contract is a set of output constraints explicitly agreed upon before generation and continuously verified during generation, typically covering dimensions such as report granularity, multimodal composition ratio and evidence grounding requirements.
MIRA's current results are based on the specific foundation model Qwen3.6-35B-A3B, and the generality of MIRA-Harness and MIRA-Data still needs to be verified on more models.
The formulation of generation contracts may currently rely on manual work or fixed templates, and how to enable agents to automatically understand and refine user intent remains a direction worth exploring.
All Facts (9)
MIRA is a multimodal text-and-image deep research agent aimed at controllable reports.
50%UnverifiedThe core component of MIRA is MIRA-Harness, which translates user requirements into a generation contract and connects four stages: evidence collection, report submission, feedback-driven revision and final delivery.
50%UnverifiedIn the evidence collection stage, MIRA's agent performs multi-step web searches, captures both textual and image evidence, and records their sources, laying the foundation for subsequent traceability.
50%UnverifiedMIRA post-trains a foundation model on MIRA-Data, enabling the model to understand the constraints of the generation contract and learn to actively use image evidence during search.
50%UnverifiedMIRA achieves state-of-the-art performance among open source agents on 7 representative multimodal search and deep research agent benchmarks.
50%UnverifiedThe text-and-image reports generated by MIRA align with user requirements, and both textual and visual evidence are traceable, allowing users to trace back the original materials underlying each conclusion, including image sources.
50%UnverifiedA generation contract is a set of output constraints explicitly agreed upon before generation and continuously verified during generation, typically covering dimensions such as report granularity, multimodal composition ratio and evidence grounding requirements.
50%UnverifiedMIRA's current results are based on the specific foundation model Qwen3.6-35B-A3B, and the generality of MIRA-Harness and MIRA-Data still needs to be verified on more models.
50%UnverifiedThe formulation of generation contracts may currently rely on manual work or fixed templates, and how to enable agents to automatically understand and refine user intent remains a direction worth exploring.
50%