Manus announced Flex on September 29 as a way to use a supported inference provider’s API key inside its existing workspace. Users choose a model and reasoning-effort setting while Manus continues to supply the agent system, tools and execution environments. The initial inference partners named in the announcement are OpenRouter, Fireworks and Modal.
The billing distinction is important: inference is charged directly by the provider, while other services and infrastructure used by a Manus task still consume Manus credits. Bringing a key therefore does not make the whole workflow free or move every charge to one bill. The announcement describes supported providers, not unrestricted compatibility with any endpoint.
Editorial take: Separating the model from the workspace may appeal to teams with an existing provider relationship or a reason to compare models on one workflow. The useful comparison is total task cost and result quality, including retries and infrastructure. A lower inference price alone does not show that the finished job is cheaper.