MiroThinker

Open-source deep research agent with top benchmark scores.

Open Source Education & Learning
Visit Website ↗ by MiroMind AI

What is MiroThinker?

MiroThinker is MiroMind AI's open-source series of RL-trained deep research agents built for complex, multi-step investigation. It pairs long-horizon reasoning with search-augmented tool use, letting agents autonomously plan, retrieve, synthesize, and verify information across extended tasks. The 1.7 release comes in 30B and 235B sizes with top-tier BrowseComp benchmark scores, plus a hosted H1 variant, and weights are available on Hugging Face. With an online demo and Apache 2.0 licensing, it is one of the strongest open options for autonomous deep research.

Key features

  • RL-trained deep research agents
  • Long-horizon reasoning with tool use
  • 30B and 235B open weights
  • Top BrowseComp benchmark scores
  • Online demo available

Pros

  • State-of-the-art open research agent
  • Open weights on Hugging Face
  • Strong benchmark results

Cons

  • Self-hosting large models needs serious hardware
  • Technical setup for the full agent stack

MiroThinker at a glance

PricingOpen Source
CategoryEducation & Learning
Platformsweb, linux
CompanyMiroMind AI
Official websitemiromind.ai
deep-researchai-agentsopen-sourcebenchmarks

How to choose the right Education & Learning tool

A learning tool earns its place by how well it teaches, not how quickly it answers. Look for step-by-step explanations, worked examples and questions that check understanding, rather than finished homework on demand — the learning happens in the attempt. Match the tool to the level: school students need guardrails, simple language and safety controls, while university and professional learners need depth, sources and subject accuracy. Feedback quality is the dividing line: specific corrections on your work teach far more than generic encouragement. If the tool will be used by children, check age limits and privacy practices carefully, and prefer tools designed to guide practice instead of replacing it.

Compare all Education & Learning tools →

MiroThinker FAQs

What is MiroThinker?

MiroThinker is MiroMind AI's open-source series of RL-trained deep research agents built for complex, multi-step investigation. It pairs long-horizon reasoning with search-augmented tool use, letting agents autonomously plan, retrieve, synthesize, and verify information across extended tasks. The 1.7 release comes in 30B and 235B sizes with top-tier BrowseComp benchmark scores, plus a hosted H1 variant, and weights are available on Hugging Face. With an online demo and Apache 2.0 licensing, it is one of the strongest open options for autonomous deep research.

What are the main features of MiroThinker?

MiroThinker offers RL-trained deep research agents, Long-horizon reasoning with tool use, 30B and 235B open weights, Top BrowseComp benchmark scores and Online demo available.

How much does MiroThinker cost?

MiroThinker is open source, so the software itself is free to download, use and self-host. Your only costs are the infrastructure you run it on and any paid services you connect to it.

What are the pros and cons of MiroThinker?

Strengths of MiroThinker include State-of-the-art open research agent; Open weights on Hugging Face; Strong benchmark results. On the downside, Self-hosting large models needs serious hardware; Technical setup for the full agent stack.