Manus Flex separates the model from the agent infrastructure

Manus Flex opens the Manus agent environment to external API keys. Users can connect their own inference provider and choose the model they want to run while keeping the same Manus projects, tools, execution environments, and automations.

The model becomes interchangeable

Until now, using Manus also meant relying on Manus to manage model selection and inference. Manus Flex separates those layers.

Users can bring their own API key from a supported provider, select a model, and configure its reasoning effort. Inference then runs through the external provider account without leaving the Manus workspace.

The rest of the environment stays in place. Projects, tools, execution environments, and automations continue to work in the same workspace. Manus also retains the agent harness responsible for planning tasks, calling tools, and coordinating execution. Two billing layers for the same task

The separation also changes how costs are handled. Model inference is billed directly by the provider connected through the API key.

Manus resources remain separate. Services and infrastructure used while executing a task continue to consume Manus credits.

Flex is therefore more than a model selector. The external provider handles inference, while Manus retains the execution layer that turns model output into actions across its environments. OpenRouter, Fireworks, and Modal join first

The Flex Inference Partner Program starts with three partners: OpenRouter, Fireworks AI, and Modal.

OpenRouter provides access to multiple model families through a unified API. Fireworks AI provides inference infrastructure for generative models, while Modal supports running AI models and workloads on its cloud infrastructure.

These integrations expand the available choices beyond large-model providers directly supported by Manus. A selected model can then be used across existing projects, automations, and agent workflows. Keeping Manus without keeping its model

The change is particularly relevant for teams that already have contracts, credits, or infrastructure with an inference provider. They can keep that commercial relationship while using Manus to orchestrate the work.

A Flex configuration is tied to a provider, a primary model, and a reasoning-effort setting. The announcement does not yet provide an exhaustive list of compatible models or detail potential feature differences between providers.

Manus Flex is available from the Manus workspace using an API key from a supported inference provider.