Quick Answer
- Built for complex, long-horizon coding tasks and high-value engineering workflows.
- Strong instruction adherence and consistency make it suitable for strict delivery requirements.
- Supports vision understanding, tool calling, and streaming for practical multimodal pipelines.
Key Parameters
- model | string | required | claude-opus-4-6 | - | Model identifier for this page (for example claude-opus-4-6).
- messages | object[] | required | - | - | Conversation messages in chronological order; role support is user and assistant.
- max_tokens | integer | required | - | >=1 | Maximum output tokens.
- system | string | object[] | optional | - | - | Top-level system prompt (do not place it inside messages).
- stream | boolean | optional | false | - | Whether to enable SSE streaming output.
- temperature | number | optional | 1 | 0-1 | Sampling temperature controlling output randomness.
- top_p | number | optional | - | 0-1 | Nucleus sampling threshold; avoid aggressively tuning with temperature together.
- stop_sequences | string[] | optional | - | - | Stop sequences to end generation on matching substrings.
- Authorization | HTTP Header | optional | - | - | Bearer auth: Authorization: Bearer <YOUR_API_KEY>.
- x-api-key | HTTP Header | optional | - | - | API key auth (common in Anthropic SDK workflows), use either this or Authorization.
- anthropic-version | HTTP Header | optional | 2023-06-01 | - | Anthropic API version header; usually set automatically by Anthropic SDK.
Common Errors
- 400 invalid_request_error: trigger=Missing fields, invalid messages schema, or mismatched parameter types.; fix=Validate model, messages, and max_tokens fields and data types.; retry=Retry only after fixing payload.
- 401 authentication_error: trigger=Missing Authorization header, invalid format, or invalid key.; fix=Verify bearer token format and key permissions.; retry=Retry after auth is fixed.
- 429 rate_limit_exceeded: trigger=Request rate, concurrency, or current quota hits upstream rate limiting.; fix=Apply exponential backoff first, then review request rate, concurrency, and quota usage.; retry=Use 1s/2s/4s backoff with jitter; if it persists, reduce submission pressure.

