DEV Community

Cover image for How to Use Claude Opus 5 in Claude Code
Hassann
Hassann

Posted on • Originally published at apidog.com

How to Use Claude Opus 5 in Claude Code

Claude Opus 5 landed on July 24, 2026, and reached Claude Code on day one. If you pay for Max, it is already your default. If you are on Pro, it is now the strongest model you can access. If you use Claude Code with an API key, it costs $5 per million input tokens and $25 per million output tokens—the same as Opus 4.8.

Try Apidog today

The part nobody warned you about: your existing setup may be working against it. Opus 5 verifies its own work without being asked, writes longer by default, and uses subagents more readily than Opus 4.8. Your carefully tuned CLAUDE.md may amplify those behaviors instead of controlling them.

This guide covers model selection, subscription-tier changes, effort settings, and four prompting edits to make on day one. You will also see how to pair it with Apidog so the model tests the API it wrote instead of assuming it works. For the full model breakdown, start with our Claude Opus 5 overview, and see Anthropic’s Claude Opus 5 announcement for the source specs.

What changes by subscription tier

Access shifted in a way that matters more than usual.

Max subscribers: Opus 5 becomes your default model. Open a session and you are already using it. Recalibrate your expectations around cost, verbosity, and delegation because your previous Claude Code habits may have been tuned for a different model.

Claude Opus 5 subscription tier interface

Pro subscribers: Opus 5 is now the highest-performing model available on this tier. Previously, the top of the stack was below the frontier tier; now Pro users can access Anthropic’s recommended default across its documentation.

API key users: Opus 5 is available through the Claude API, Amazon Bedrock (anthropic.claude-opus-5), Google Cloud, Microsoft Foundry, and GitHub Copilot. Pricing is $5 input and $25 output per million tokens, unchanged from Opus 4.8 and half of Fable 5’s $10/$50 rate. See the full table, including caching and batch rates, in our Opus 5 pricing breakdown.

Claude Opus 5 pricing comparison

One spec worth knowing before you start: Opus 5 has a May 2026 knowledge cutoff, the freshest of any current Claude model. Fable 5 and Sonnet 5 both stop at January 2026. For coding work involving libraries and APIs released in the last quarter, that five-month difference can matter.

Select Opus 5 in Claude Code

Use the /model command inside a Claude Code session:

/model
Enter fullscreen mode Exit fullscreen mode

Choose Claude Opus 5 from the model picker.

To switch directly without opening the picker, pass the model ID:

/model claude-opus-5
Enter fullscreen mode Exit fullscreen mode

The API model ID is exactly:

claude-opus-5
Enter fullscreen mode Exit fullscreen mode

Use that string when scripting Claude Code, setting project defaults, or writing rules that reference the model. Anthropic’s model overview documentation lists platform-specific IDs for Bedrock and Google Cloud.

Opus 5 supports 1M tokens of context by default and at maximum, with no beta header or long-context price premium. Maximum output is 128K tokens on the Messages API.

Large repositories can fit in context, but avoid sending the entire repository on every turn. Scope context to the files, logs, and specs relevant to the task. For more commands, see our Claude Code cheatsheet.

How effort behaves

The cost-versus-capability control is the output_config.effort API parameter. It has five levels:

low
medium
high
xhigh
max
Enter fullscreen mode Exit fullscreen mode

Consumer interfaces expose this as an effort selector.

Start with the default: high

If you do not specify effort, Opus 5 uses:

high
Enter fullscreen mode Exit fullscreen mode

Do not assume it defaults to medium or an adaptive setting without a floor.

Re-run your effort evaluation

Effort levels are recalibrated for Opus 5. Do not carry over Opus 4.8 settings without testing them against your own tasks.

In particular, low and medium are substantially stronger than on earlier Opus models. Tasks you would not have trusted to medium on Opus 4.8 may now work at a lower cost.

For coding and agentic tasks, Anthropic recommends starting with:

xhigh
Enter fullscreen mode Exit fullscreen mode

Then run an evaluation sweep across representative tasks:

  1. Pick tasks with known expected outputs.
  2. Run each task at medium, high, and xhigh.
  3. Record success rate, latency, and token usage.
  4. Use the lowest effort level that reliably passes.

Avoid the thinking and effort conflict

This combination returns a 400 error:

{
  "thinking": { "type": "disabled" },
  "output_config": {
    "effort": "xhigh"
  }
}
Enter fullscreen mode Exit fullscreen mode

The same applies to max effort.

Claude Code manages request shape in interactive sessions, so you are unlikely to hit this manually. You can hit it when scripting Claude Code or calling the SDK directly.

Use one of these fixes:

  • Keep thinking enabled and use the effort level you need.
  • If thinking must be disabled, cap effort at high.

Keep thinking enabled when possible. With thinking disabled, Opus 5 can occasionally emit tool calls as plain text instead of executable calls, which can pollute later turns in an agent loop. Internal XML tags may also appear in visible output.

Control cost by lowering effort rather than disabling thinking. For a practical evaluation process, see our effort parameter guide and the Opus 4.8 migration guide.

Four prompting edits to make on day one

These changes will save the most cost and frustration. They are based on Anthropic’s Opus 5 prompting guide.

1. Delete verification instructions

If your CLAUDE.md includes instructions such as:

  • “Always double-check your work.”
  • “Verify the change before reporting.”
  • “Confirm the tests pass and then confirm again.”

Remove them.

Opus 5 verifies its work without being asked. Adding extra verification instructions can cause over-verification: rereading files, rerunning checks, and consuming tokens to reconfirm completed work.

Audit both project-level and global CLAUDE.md files:

grep -RniE "verify|double-check|confirm" ~/.claude ./CLAUDE.md
Enter fullscreen mode Exit fullscreen mode

Remove instructions that only restate generic verification behavior. Keep concrete requirements, such as “run npm test before reporting,” when they define a project-specific acceptance check.

2. Cap subagent spawning

Opus 5 delegates to subagents more readily than Opus 4.8. That helps with large parallel research tasks, but it can increase cost for work that one agent could finish.

Add a delegation policy to your project instructions:

## Delegation

Do not spawn subagents for tasks that a single pass can finish.
Use at most two parallel subagents. If a task looks like it needs
more, describe the split and ask before spawning.
Enter fullscreen mode Exit fullscreen mode

The goal is not to disable delegation. It is to make the model justify parallel work before creating more agents. If you are building a structured subagent workflow, see our guide to creating Claude Code subagents.

3. Prompt for conciseness explicitly

Opus 5 produces longer responses, summaries, explanations, and written deliverables than Opus 4.8 by default.

Lowering effort does not solve this. Effort controls reasoning depth, not visible response length. Dropping to medium can reduce reasoning while leaving you with similarly long output.

Set output length in your instructions instead:

## Response style

Report results in under 150 words unless I ask for detail.
No summary of what you just did if I can see the diff.
Skip preamble. Lead with the answer.
Enter fullscreen mode Exit fullscreen mode

Use an explicit word count, bullet limit, or required response format for tasks where output length matters.

4. Constrain narrow tasks explicitly

Opus 5 expands scope more readily and narrates corrections as it works. If you ask it to fix one function, it may decide the surrounding module also needs attention.

For a narrow change, define the boundary:

Fix the null check in parseConfig and change nothing else.
Do not refactor adjacent functions.
Run only the relevant tests.
Enter fullscreen mode Exit fullscreen mode

This is more reliable than asking only:

Fix the null check in parseConfig.
Enter fullscreen mode Exit fullscreen mode

For more copy-paste patterns, see our guide to prompting Opus 5.

Build an API, then let Opus 5 test it

Claude Code can scaffold routes, but it cannot confirm that a running endpoint follows the contract unless you connect it to an external verification loop.

Use this workflow:

  1. Design endpoints, schemas, and examples in Apidog.
  2. Create a mock server that returns realistic responses.
  3. Connect Claude Code to the API specification through the Apidog MCP server.
  4. Have Opus 5 implement against the actual schema rather than inferring one.
  5. Run an API test scenario after implementation.

For example:

apidog run --access-token $APIDOG_TOKEN \
  --project-id $PROJECT_ID \
  --test-scenario "user-signup-flow"
Enter fullscreen mode Exit fullscreen mode

Failed assertions become output that the model can inspect and fix. That is a verification loop grounded in an external contract, which is stronger than self-verification alone.

The same setup helps when calling the Claude API directly. With Apidog, you can:

  • Send claude-opus-5 requests.
  • Inspect SSE streaming events.
  • Review tool-call payloads.
  • Store API keys in environment variables.
  • Compare the same prompt at two effort levels.
  • Inspect the usage block to confirm cache hits.

Download Apidog to set up a local project.

The honest ceiling

Opus 5 is not the top of the Claude stack.

Fable 5 still holds Anthropic’s “most capable widely released” designation. Opus 5 also trails Mythos 5 on cybersecurity exploitation and autonomous biology research, according to Anthropic.

If your Claude Code work is in either area, Opus 5 is not the right ceiling. Our Mythos-class explainer covers what that tier is for.

On benchmark claims, Anthropic reports that Opus 5:

  • More than doubles Opus 4.8’s Frontier-Bench v0.1 score.
  • Reaches roughly 3× the next-best model on ARC-AGI 3.
  • Comes within 0.5% of Fable 5’s CursorBench 3.2 peak at half the price.

These are vendor-run numbers. As of July 25, 2026, none have been independently reproduced. Treat them as a reason to run your own evaluation, not as a settled result.

The practical summary: Opus 5 offers frontier-class capability at half the frontier price, with a named ceiling above it. That is useful, but it is not the same as claiming it is the best model available.

When to use something else

Use Opus 5 for most Claude Code work:

  • Agentic coding
  • Multi-file refactors
  • Long-horizon tasks
  • Workflows where the model runs commands and iterates on output

Switch when the workload is high-volume and cost-dominated, where Sonnet 5 can handle standard editing at a lower rate. Use Fable 5 or Mythos 5 for the hardest frontier tasks in their respective domains.

Switching models is one command:

/model claude-sonnet-5
Enter fullscreen mode Exit fullscreen mode

Use Sonnet 5 for lower-cost work, then return to Opus 5 for harder tasks:

/model claude-opus-5
Enter fullscreen mode Exit fullscreen mode

If you previously used Sonnet 5 in Claude Code or Fable 5 in Claude Code, the workflow is familiar.

Practical tips

  • Audit CLAUDE.md first. Remove generic verification instructions, add a response-length constraint, and cap delegation.
  • Re-sweep effort. Your Opus 4.8 settings are stale; low and medium may now handle tasks you previously reserved for higher effort.
  • Leave thinking on. Control cost with effort instead of disabling thinking.
  • Scope context. A 1M-token window is room for large tasks, not a reason to send the full repository every turn.
  • Verify externally. Self-verification is useful; a failing test against a real API contract is better.

FAQ

Is Claude Opus 5 available in Claude Code?

Yes. It has been available since its July 24, 2026 release. Use /model and choose Claude Opus 5, or run:

/model claude-opus-5
Enter fullscreen mode Exit fullscreen mode

It is the default for Max subscribers and the highest-performing model available to Pro subscribers.

Do I need to change my CLAUDE.md for Opus 5?

Probably. Anthropic’s prompting guide says verification instructions carried over from earlier models can cause over-verification because Opus 5 already verifies work unprompted.

Remove generic verification prompts. Add an explicit conciseness requirement and a subagent cap.

Why did lowering effort not make responses shorter?

Effort controls thinking depth, not visible output length. Lowering effort reduces reasoning tokens but may leave response length roughly unchanged.

Ask for a shorter output explicitly in the prompt or CLAUDE.md.

What causes a 400 error when I set effort to xhigh?

This combination is invalid on Opus 5:

{
  "thinking": { "type": "disabled" },
  "output_config": {
    "effort": "xhigh"
  }
}
Enter fullscreen mode Exit fullscreen mode

The same restriction applies to max. Enable thinking or cap effort at high. See the migration guide for other Opus 4.8 breaking changes.

Can I use Opus 5 in Claude Code on a free plan?

Opus 5 is available on Pro and Max subscriptions and through the API. Pro users access it as the top model for that tier, while Max users get it as the default.

Our Opus 5 free access guide covers the available paths and the cheapest paid options.

Top comments (0)