Manus vulnerable to model vendor capture

This is the core platform risk for Manus, because its product can be copied from above by the model vendors it depends on and squeezed from below by the infrastructure vendors it pays. Manus sells a credit based agent that researches on the web, uses tools, runs code, and fills forms, but OpenAI has already merged browser use, research, and code execution into ChatGPT agent, while Anthropic has added web search and multi agent style capabilities around Claude. That makes Manus vulnerable on both cost and access.1 2 3 4 5

• Manus is not just buying raw model tokens. It also relies on outside components for deep research and web retrieval. Internal research describes Parallel as the backbone of Manus deep research, which means a meaningful part of the user experience sits on vendors Manus does not control. If any of those vendors raise prices, rate limit usage, or launch their own full agent, Manus margin gets hit immediately.1 4 6
• The threat is concrete because the labs are moving down the stack into the exact workflow Manus packages. OpenAI introduced Operator in January 2025, then folded Operator, deep research, and terminal based code execution into ChatGPT agent in July 2025. Anthropic now offers web search in Claude, and internal Manus research notes Claude Code expanding with web search and parallel sub agents. The labs are turning base models into finished products.1 2 3 5 7
• The closest comparable is Perplexity, which is making the opposite bet. Perplexity Computer is built as a multi model orchestration layer and prices with credits on top of subscriptions, explicitly betting that workflow design and accumulated user context can stay valuable even as labs copy individual features. Manus likely needs the same escape route, owning a workflow or data layer that survives model substitution.7 8

This market is heading toward fewer standalone agent winners and more pressure to own either the full stack or a defensible workflow niche. For Manus, the path forward is to reduce dependence on any one lab, deepen proprietary integrations and context, and become the place where work actually happens, not just the interface sitting on top of someone else's agent stack.1 4 7 8