How to Use Claude Opus 4.5 (Dec 2025)

1) Use Opus 4.5 in the Claude App

1. Sign in at claude.ai (Pro/Max) or use Team/Enterprise.
2. Start a new chat and pick Opus from the model selector (Opus 4.5 is the current Opus).
3. Paste your task or attach files/images; keep inputs under ~200K tokens total.
4. For coding: paste the relevant files/snippets and be explicit about files to change.

2) Call Opus 4.5 via the Claude API

Model ID: claude-opus-4-5-20251101
Endpoint: https://api.anthropic.com/v1/messages
Pricing: $5 / 1M input tokens; $25 / 1M output tokens (API). Prompt caching and batch routes lower costs for repeated prompts.

Example (cURL):

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-opus-4-5-20251101",
    "max_tokens": 800,
    "messages": [
      {"role": "user", "content": "Review this function for edge cases:\n```python\n...code...\n```"}
    ]
  }'

Tips:

• Keep total tokens (prompt + response) within 200K.
• Use system to set style/constraints; keep it stable to benefit from prompt caching.
• For tools: include a tools array and let Claude return tool_use calls per the Messages API.

3) Use Opus 4.5 on Cloud Partners

• Amazon Bedrock: Select the Claude Opus 4.5 model ID in Bedrock; billing via AWS.
• Google Vertex AI: Choose the Opus 4.5 model from Vertex’s catalog; billing via GCP.
• Microsoft Foundry: Select Opus 4.5 where available; billing via Azure.

Check each provider’s rate card; prices differ from the Claude API.

4) Good Prompts and Practices (from Anthropic docs)

• Be explicit: state goals, constraints, and the files or APIs involved.
• Provide structure: use bullet lists or numbered steps to reduce ambiguity.
• Keep context lean: trim logs, include only relevant file sections; stay under 200K tokens.
• For multi-step work: ask for a short plan first, then approve/iterate.
• Use JSON/tool calls when you need structured actions instead of free text.
• Cache repeated system prompts or boilerplate to cut cost and latency (prompt caching).

5) When to Pick Opus 4.5 vs Sonnet 4.5

• Use Opus 4.5 for hardest tasks: multi-file refactors, deep debugging, long-context analysis, or high-stakes reasoning.
• Use Sonnet 4.5 for day-to-day coding and chat when cost/latency matter more than peak capability (also 200K context, lower price).

6) Quick Checklist

• [ ] Confirm access (Claude Pro/Max/Team/Enterprise or API key).
• [ ] Use claude-opus-4-5-20251101 as the model.
• [ ] Stay within 200K tokens total.
• [ ] Set max_tokens to cap spend.
• [ ] Consider prompt caching/batch for repeated prompts.
• [ ] Prefer tools/JSON for structured tasks.

Sources (December 2025)

• Claude pricing/models: https://claude.com/pricing
• Claude API reference: https://docs.anthropic.com/docs/en/api/overview
• Claude tool use & best practices: https://docs.anthropic.com/docs/en/build-with-claude/working-with-messages
• Partner availability: Bedrock/Vertex/Foundry provider catalogs

Want more? See our other Claude guides and API quickstarts in the directory.