Use an Ollama model route with Claude Code and inspect token usage

Choose the local or cloud route

Thanks to Tigerless Lab for sponsoring this video.

The video uses GLM 5.2 through Ollama, showing both local and cloud approaches. It is not proof that every open-weight model or current plan works with the same command.

1. Check the model route and machine requirements for the approach you intend to use.
2. Follow the recorded launch sequence rather than inventing an endpoint configuration.
3. Verify the actual model and route before assigning repository changes.
4. Keep the working client configuration recoverable.

Try a small, checkable task

Luke presents a judgement about jobs he would hand to a cheaper model and notes a /model switching quirk. That is specific to the setup shown in July.

1. Choose a small task with an observable expected result.
2. Inspect the generated change and run its relevant check.
3. When following the recording, preserve the demonstrated model-switching workaround instead of assuming a command changed the active route.
4. Record where the result needs premium-model or human follow-up.

Inspect costs separately from the headline

The sponsor segment introduces the local cost-xray tool for seeing usage by prompt, tool and MCP server. Historical plan prices are not a current comparison.

1. Open the retained cost-xray repository for its actual installation instructions.
2. Inspect the usage categories relevant to the task, including fresh and cached tokens.
3. Compare the complete route and correction effort before deciding what to offload.
4. Check current local/cloud usage terms rather than relying on the July price discussion.

The takeaway

The useful pattern is a verified model route and a bounded task with inspectable usage. The recording does not establish that any open model is a drop-in replacement for every Claude workflow.