Anthropic API

The free credit can be used for the Claude API, Agent SDK, Managed Agent and Playground.

Halogen is precisely optimized for AMD's Strix Halo chip (GFX1151) plus one model family; it has no browser UI and is a pure API backend, natively implements the Anthropic API, and Cloud Code can connect directly to local models.

A beta value is a request header marker in the Anthropic API used to enable experimental features, typically sent with requests in the form anthropic-beta.

Claude Code supports forwarding requests via configuration to other services compatible with the Anthropic API format, as described in the official documentation.

GitHub Copilot's three tiers are priced at $10, $39, and $100, including $15, $70, and $200 of AI credits respectively, with Claude metered at the same rates as the Anthropic API.

The API version (interface contract) and the model version (model ID) are two independent dimensions: the former constrains interface behavior, and the latter determines inference capability.

The Anthropic API uses a date-based versioning mechanism: each API version is named after its release date and is specified via the anthropic-version request header or SDK configuration, with semantics frozen after release.

The list spend limits endpoint previously returned limits at the organization or key dimension.

When upgrading, the API version number should be pinned in code or environment variables, and each item checked against the migration guide.

The server decides which format to return based on the version number declared by the client.

Anthropic's API uses date-identified version numbers, explicitly specified via a request header (usually anthropic-version).

Prompt Caching is a mechanism provided by large-model service providers such as the Anthropic API: when the request prefix is exactly the same as that of a previous request, the server reuses the already-computed KV Cache, the cache-hit portion is billed at 0.1 times the input price, and the first cache write is billed at 1.25 times.

The cost weights adopted by the author are: input 1, cache write 1.25, cache read 0.1, output 5.

Whether web search is available depends on the application, plan, and settings; API calls have independent tool configuration.

Prompt caching requires the request prefix to match byte for byte, and caches usually have a time to live (such as 5 minutes to 1 hour), after which they must be written again.

The Anthropic API has quotas on requests per minute and input tokens per minute, and an agent stuck in a loop may hit the 429 rate limit within seconds, which then affects other tasks under the same account.

In the Anthropic API, context overflow manifests as a 400 error.

Prompt caching is a capability of the Anthropic API: when the prefix portion of a request is exactly identical to the previous request, the server can reuse the already-computed intermediate state and skip redundant prefill computation.

529 Site is Overloaded is not a standard HTTP status code but an extension status code proprietary to the Anthropic API, used to indicate that the server is temporarily unable to handle requests due to a momentary surge in traffic.

Anthropic uses a token-based API pricing model, where stronger models (such as the Opus series) have higher unit prices.