Manus
General AI agent that autonomously executes multi-step tasks across research, code, and the web.
About
Manus, built by Butterfly Effect (Monica), launched in invite-only beta in March 2025 as a self-described general AI agent. It runs a multi-agent architecture on Anthropic's Claude plus fine-tuned Alibaba Qwen models, and its signature “Manus's Computer” interface lets you watch the agent work and intervene mid-task. The launch demo drew over a million views in under a day, invitation codes traded on secondary markets for thousands of dollars, and it posted state-of-the-art GAIA benchmark scores at release.
The honest picture: the credit-based pricing has been criticized as opaque — similar-looking tasks can burn very different amounts of credit — and Trustpilot sentiment skews negative on billing and refunds. It still struggles with paywalled sites and CAPTCHAs, and per-task cost runs roughly $2. In December 2025 Meta agreed to acquire the company for a reported $2-3 billion, so the product's long-term direction is now in someone else's hands.
What it can do
- Autonomous multi-step task execution
- “Manus's Computer” live-intervention interface
- Web browsing, code execution, and file handling
- Artifact publishing to shareable URLs
- Multi-agent architecture (Claude + fine-tuned Qwen)
Setup
Sign up at manus.im. Free tier includes starter credits with a daily refresh; paid plans start at $20/month.
Prerequisites
A Manus account. Complex multi-step tasks consume significant credits, so check estimates before long runs.
Use cases
Works with
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 51; Trending score 92/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.