Manus hits $450M run rate amid Meta deal fallout

Manus reached a $450M annualized revenue run rate in June 2026.

Manus has moved from invite-only curiosity to a company with real money attached to it. Sacra pegs its annualized revenue at $450 million in June 2026, up from $127 million earlier in 2025, while the company also sits in the middle of a messy ownership story involving Meta, Chinese regulators, and a possible buyback.

The bigger story is that Manus is trying to be two businesses at once. It sells a consumer and team subscription product today, but it is also building an SDK so other companies can embed Manus workflows inside their own apps.

Revenue is climbing faster than the product story

Manus launched paid plans in March 2025 after spending 2024 as a free, invite-only beta. That matters because the company went from hype to monetization in a very short window, and the numbers suggest users were willing to pay once the gates opened.

Sacra says Manus hit a $125 million revenue run rate in December 2025, then crossed $450 million annualized revenue by June 2026. That is a sharp jump for a company that only introduced subscriptions in 2025.

The pricing structure helps explain the ramp. Manus starts at $19 per month for basic plans and reaches $199 per month for premium tiers, with usage-based credits layered on top. For individuals, the Manus Starter plan costs $39 per month and includes 3,900 credits, while the Pro plan costs $199 per month and includes 19,900 credits.

• March 2025: paid plans launched
• December 2025: $125M run rate
• June 2026: $450M annualized revenue
• Starter plan: $39 per month, 3,900 credits
• Pro plan: $199 per month, 19,900 credits

The company is selling autonomy, not chat

Manus is an autonomous AI agent platform. In practice, that means a user types a task in plain English, and Manus spins up a cloud virtual machine with tools like browsers, code interpreters, office apps, and design software to finish the job.

That makes the product feel closer to a worker than a chatbot. If you ask it to analyze 100 sneakers and produce a sortable spreadsheet, it breaks the task into steps, visits websites, pulls pricing data, normalizes the fields, and returns both chat output and downloadable files.

The company says it uses Anthropic models, especially Claude, plus its own context engineering for prompt scaffolding and memory management. That dependency is important because it gives Manus speed to market, but it also leaves the company exposed to pricing changes and product overlap from model providers.

“The general-purpose model is a very powerful platform, but it is also a very powerful competitor.” — Dario Amodei, Anthropic CEO, in a 2024 interview with The Verge

That quote fits Manus neatly. The company is building on top of foundation models while trying to keep enough product distance to avoid being swallowed by them.

Manus has also pushed the product past simple task completion. Wide Research, launched in July 2025, lets up to 100 sub-agents work in parallel. Manus 1.5, released in October 2025, cut average task completion time from 15 minutes to under 4 minutes. Manus 1.6 Max, released in December 2025, added Design View and lifted user satisfaction by more than 19.2% in double-blind testing.

Funding, ownership, and geopolitics are all part of the same story

Manus raised $75 million in a Series B led by Benchmark in April 2025, which took total funding to $85 million. Earlier investors include Tencent, HongShan, and ZhenFund.

Then the ownership story got complicated. Meta announced an acquisition in December 2025 at a reported valuation of more than $2 billion. China blocked the deal in late April 2026, citing national security concerns. By June 11, 2026, Meta had completed an operational split from Manus, cutting off data sharing and internal access.

That left Manus in limbo. The Information reported on June 18, 2026 that early Chinese backers were pursuing a buyback at roughly the same $2 billion valuation Meta paid, with a structure that could leave Manus as a joint venture and possibly set it up for a Hong Kong IPO.

• January 2023: $10M Series A
• April 2025: $75M Series B led by Benchmark
• December 2025: Meta acquisition announced at more than $2B
• April 2026: China blocked the deal
• June 2026: operational split completed

Manus is trying to become infrastructure for other apps

The most interesting part of Manus may be what happens after the subscription business. The company is building an agent-as-platform SDK so other companies can embed Manus workflows directly into their products.

That changes the economics. Instead of selling one subscription per user, Manus could charge developers and collect revenue from every downstream user who interacts with a partner app. It also moves the company closer to an app-store style model, where distribution matters as much as the core agent.

Manus is already testing vertical packs for finance, merchandising, and legal research. Those bundles add domain prompts, specialized integrations, and compliance features, which is exactly where generic agent tools often fall short.

There is also a partnership angle. Manus is a launch partner for Microsoft Agent 365, which means its workflows can run inside Microsoft 365 after enterprise setup. It also has a partnership with Alibaba Cloud’s Qwen team, giving it access to multimodal models and cloud resources.

That matters because the company is not trying to win every layer of the stack. It is trying to sit inside the stack businesses already use.

Competition is coming from both sides

Manus is squeezed by two very different groups. On one side are model giants like OpenAI, Google, Microsoft, and Anthropic, all of which can bundle agent features into broader platforms. On the other side are open-source frameworks like LangChain and AutoGen, which let teams build custom workflows without paying a hosted vendor.

That pressure shows up in pricing. Open-source tools require more engineering work, but they remove subscription fees and give teams total control. Commercial products have to justify themselves with speed, reliability, and better workflow design, not just the fact that they can run an agent.

Specialized rivals matter too. Cognition focuses on software engineering with Devin, while Adept built ACT-2 around UI control. Those companies can go deeper in one workflow, which makes them harder to dismiss than generic assistants.

• OpenAI, Google, Microsoft, and Anthropic can bundle agent features into larger products
• LangChain and AutoGen let teams build custom agent stacks
• Cognition and Adept target narrower, higher-value workflows
• Manus counters with subscriptions, SDK plans, and vertical packs

What to watch next

Manus has already proved it can turn hype into revenue, but the next test is whether it can keep growing while its dependencies and ownership structure stay messy. If the SDK gains traction, Manus could look less like a consumer AI tool and more like middleware for agent workflows.

The near-term question is simple: can Manus keep converting product momentum into durable distribution before the bigger model companies copy the parts that matter? If it can, the subscription business is only the first chapter.