A real cost problem when mixing Claude Pro with local models
A Claude Pro subscriber wants to use a local inside the same workflow as Claude models. The current setup runs a local MCP server with Ollama and the model.
The goal is to call Qwen3-Coder the same way Haiku can be called, and also let it work as a sub agent when needed. Sonnet and Opus should still remain available in the same setup.
The problem is that the available methods seem to rely on the Claude API, which means paying by usage on top of the Claude . The main question is whether Claude Pro can be used together with both a local model and Anthropic cloud models in one practical setup.
Key points
- The desired setup keeps and Opus available while adding a local .
- The local stack already includes an MCP server, Ollama, and .
- Qwen3-Coder should be callable like Haiku and usable as a sub agent.
- Known methods appear to require the Claude API, which adds usage-based charges.
- The practical issue is whether Claude Pro can power this mixed local-and-cloud workflow.