Draft
Add a Second Frontier Model to Claude Code with a Documented Fallback Path
Task
What you will complete
Configure a second model and complete the same task via the fallback after a simulated failure.
Success criteria
- Claude Code has a second frontier model configured, and the fallback path is written down before the failure is simulated.
- The primary path fails on purpose, and the documented fallback is the path that then runs.
- The same task completes on the fallback, with both attempts kept for the evidence section.
What this is not
Not a benchmark of the two models, and not a general introduction to Claude Code.
Prerequisites
- Claude Code already completing one task on a primary model.
- Access to a second frontier model you can name in Claude Code's configuration.
- A way to force the primary path to fail without changing the task.
Environment
Not recorded. A verified revision will list the operating system, the tool builds, and the model identifiers that were present for the run.
1. Name the task the primary model already completes
Choose a task Claude Code already finishes on the primary model. Write the success check in one sentence so the fallback has to meet the same check, not a weaker one.
2. Add the second model
Configure the second frontier model in Claude Code using the mechanism you will document. Write the model ids in your notes. This page will not pin them until the run.
3. Write the fallback path before you need it
Describe, in the order Claude Code will follow, what happens when the primary path fails: the signal you treat as failure, the second model, and how you can tell which model produced the result.
4. Simulate the primary failure
Force the primary path to fail without changing the task. Confirm the fallback you wrote down is the path that actually ran. If a different path ran, the documentation is wrong and the tutorial is not done.
5. Complete the task on the fallback and keep both attempts
Finish the task through the fallback. Save the failed primary attempt, the fallback attempt, the model ids, and the configuration. The evidence section stays empty until those exist.
Evidence it actually ran
This is the shape of the stamp. It has not been earned, and the example date that follows is a placeholder, not a run: Ran on YYYY-MM-DD (Asia/Shanghai) with the pinned versions.
- Command output
- None recorded. A verified revision will quote the output from the run, not a sample written in advance.
- Tests, CI, or Inspector
- Not run for this draft.
- Repository or config
- No link recorded.
Failure modes
Anticipated only. None of these has been checked against a run of this tutorial.
Unchecked
The simulated failure never triggers the fallback
The primary path can error in a way your notes do not treat as the signal to switch.
Unchecked
The fallback model runs and cannot use the same tools
A second model that answers in prose while dropping tools has not completed the task.
Unchecked
You cannot tell which model produced the result
Without a record of the model id on each attempt, the fallback is not documented.
Related
Other tutorials in Integrate.
Connect a Local MCP Server to Cursor and Use It to Finish a Real Repo Task
Start a local MCP server, connect it in Cursor, and complete a repo task that requires that tool.
Not yet verified
1 min read
Build a TypeScript MCP Server for a Real API, Test It with Inspector, and Call It from Claude Code
Implement the server, pass Inspector, then call it from Claude Code to finish a real API task.
Not yet verified
1 min read
Call Qwen4-Max Through the OpenAI-Compatible API and Finish a Tool-Calling Task
Call qwen4-max from the OpenAI SDK on Alibaba Cloud Model Studio in Beijing by changing only api_key, base_url, and model, stay on that model's official free quota with 免费额度用完即停 / Free Quota Only turned on, then finish a tool-calling job: a checked dependency report for an npm project.
Not yet verified
15 min read
Call the Official Hosted Qwen3.8-27B API on Alibaba Model Studio
Call the official hosted qwen3.8-27b API on Alibaba Cloud Model Studio in Beijing, stay on that model's own free quota with 免费额度用完即停 / Free Quota Only turned on, and extract five fields from a fixed pytest failure into JSON that a local checker can verify.
Not yet verified
15 min read