​

Anthropic Claude Opus 5

​

Migration quickstart

1. Change the model ID to anthropic/claude-opus-5 without changing the rest of the request.
2. Re-run your production prompt, tool, and structured-output evaluations.
3. Rebaseline latency, token use, and task cost for long-context and high-effort workloads.
4. Canary the model behind a fallback to Claude Opus 4.8 before changing your default route.

{
  "model": "anthropic/claude-opus-5",
  "messages": [
    {
      "role": "user",
      "content": "Review this implementation plan and identify the highest-risk assumptions."
    }
  ]
}

​

What’s new

• the Phaseo model ID is anthropic/claude-opus-5
• the context window is 1M tokens
• the maximum output is 128K tokens
• the model is positioned for longer-running reasoning, coding, and autonomous tool workflows
• pricing remains aligned with Claude Opus 4.8

​

What to test

​

Prompts and outputs

• instruction following on your longest system prompts
• structured-output and schema pass rates
• response length and stopping behavior
• regressions in prompts tuned specifically for Claude Opus 4.8

​

Tools and agents

• tool selection and argument quality
• recovery after tool errors or partial results
• long-horizon completion rate
• permission and approval boundaries in high-autonomy workflows

​

Operations

• latency and token use at each effort level you expose
• context-window usage on long documents and agent traces
• cost per successful task
• fallback behavior when the primary route times out or fails

​

Safe rollout

1. Keep Claude Opus 4.8 available as a fallback during evaluation.
2. Shadow representative production traffic before serving user requests.
3. Canary by workload rather than moving every Opus route at once.
4. Promote only after quality, latency, cost, and tool-safety checks meet your targets.

​

Sources

• Claude Opus 5 announcement
• Claude Opus 5 system card