Back to directory

MiroThinker

An open-source deep research agent optimized for complex research and prediction tasks, scoring highly on the BrowseComp and BrowseComp-ZH benchmarks.

MiroThinker

MiroThinker is a deep research agent optimized for research and prediction tasks. Its latest models score highly on the BrowseComp and BrowseComp-ZH benchmarks, and a smaller MiroThinker-1.7-mini variant sets an open-source state of the art on BrowseComp-ZH while using fewer parameters than earlier full-size releases.

An earlier release, MiroThinker-v1.0, introduced interactive scaling as a way to improve performance by supporting a 256K context window and up to 600 tool calls in a single task, and is available in 8B, 30B and 72B parameter sizes. Online, the hosted MiroThinker product supports generating, previewing and sharing research reports, and accepts uploaded documents in formats such as PDF, Word, PowerPoint, Excel and images.

The project provides benchmark evaluation scripts and trace collection tooling alongside the agent itself, and is released under the Apache 2.0 license.

More in Models

See all 405
  • mini-AGI

    github.com

    A continual learning language model that trains from scratch on a single 8 GB VRAM GPU, assembling its own architecture and adapting its size to the hardware and data available rather than shipping as a fixed pretrained model.

  • An open-source contrastive-embedding pipeline that maps technical CVE vulnerability descriptions to formal cybersecurity compliance controls in NIST SP 800-53 and CMMC, published with its model generation, fine-tuning and evaluation code.

  • peekaboolean

    github.com

    A small vision-language model, fine-tuned with LoRA on a 500-million-parameter backbone, that answers typed multiple-choice, ordinal-score and yes/no questions about an image with calibrated probabilities, in about 400 milliseconds on a laptop.