MiroThinker

What it does

MiroThinker is a deep-research agent and model family for complex research and prediction tasks. Its open-source MiroThinker-1.7 models come in 30 B and 235 B parameter scales, with a stated 256 K context window for long-horizon, multi-step analysis. In the documented agent setup, a user supplies a question; the agent can search the web, extract and summarize page content, execute code when needed, and return an answer with sources. Its separately hosted online experience supports generating, previewing and sharing reports, and uploading PDFs, office documents and images.

The product is intended for research and prediction work, including researchers who want to run or evaluate the agent. It is available through a hosted web app or as a self-hosted Python framework. Self-hosting requires Python 3.10+, uv, a model server, and API keys for selected tools. Search, scraping, code execution, summary-model, and inference services may require separate credentials and charges. The model is text-only; multimodal tasks require additional processing tools. The repository is licensed under Apache-2.0.

What it can do

• Long-horizon, multi-step research and reasoning
• Web search and webpage content extraction
• Code execution through configured tools
• Research report generation, preview, and sharing
• Upload PDF, office documents, and image files
• Benchmark evaluation and agent interaction trace collection

Works with

• Serper API
• Jina
• E 2 B
• OpenAI API

Before you choose

• Self-hosted use depends on external tool and model services; their credentials and charges may be separate from the open-source code.
• The model is text-only; image, audio, or video tasks require additional preprocessing or optional tools.
• Hosted-service pricing is distinct from self-hosting, which requires compute and configuration.

Listed under

• Productivity
• Research
• Analysts
• Deep research
• Developers
• Open source
• Reasoning
• Research agent

Get started

The hosted app links to its sign-in/start page; the repository also documents installing and configuring the framework.

1. For self-hosting, install Python 3.10+ and uv, clone the MiroThinker repository, and run uv sync in apps/miroflow-agent.
2. Create and edit .env with credentials for the tools and summary model you choose; the documented minimal setup names Serper, Jina, and E 2 B keys.
3. Serve a MiroThinker model using a documented inference framework, configure the model endpoint in the agent command, and submit a question.

The hosted service lists 100 credits every 24 hours on Free and 20,000 credits per month on Pro. Starter Boost provides 4,000 credits for 365 days as a one-time purchase. Self-hosted operation and external model/tool services may have separate costs. These are hosted-service entitlements, not fees for downloading the open models or framework. Check the current hosted billing selector before purchase.

Source: miromind.ai/pricing, checked 1 Nov 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 1 Nov 2026. Each source supports part of the overview; availability and prices can change.

• GitHub - MiroMindAI/MiroThinker: MiroThinker is a deep research agent optimized for complex research and prediction tasks. Our latest models, MiroThinker-1.7, achieves 74.0 and 75.3 on the BrowseComp and BrowseComp Zh, respectively. · GitHub ↗github.com
• MiroMind | General Purpose Solver ↗www.miromind.ai
• MiroMind | General Purpose Solver ↗www.miromind.ai
• MiroMind | General Purpose Solver ↗www.miromind.ai

Company website

miromind.ai: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

manual_review

Last checked 1 Nov 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.