Skip to content
ClaudeOpus

Anthropic-compatible API

Every Claude model. One base URL.

Point the Anthropic SDKs, Claude Code or Cursor at ClaudeOpus and keep your code as it is — every key's 5-hour window reads out per model, to a tenth of a percent.

Get a keyRead the docs

Keys are issued by your admin or reseller — sign in to see yours.

Base URLhttps://api.claudeopus.pro
Example keyDemo data
Active

Demo readout with illustrative data, not a real key: a 5-hour window at 63.4% used — Fable 24.1%, Opus 18.6%, Sonnet 14.9%, Haiku 5.8% — resetting in about two hours, above a list of its most recent requests.

Illustrative readout — not a live key.Read a real key's window
Context
1Mtokens
Fable 5.1 · Opus 5 · Sonnet 5
Usage window
5 hrolling
Per key, read out per model
Models
14IDs
On one base URL
Prompt caching
Free
Cache reads and writes never count against the window

01. Lineup

Ask for any model by its ID.

The four current models, flagship first. Each also answers to its family's short alias — fable, opus, sonnet or haiku.

  • Claude Fable 5.1

    Flagship

    The most capable model served: demanding reasoning and long-horizon agentic work.

    Model ID
    claude-fable-5-1
    Alias fable
    Context window
    1,000,000tokens
    Also listed as claude-fable-5-1[1m]

  • Claude Opus 5

    Deep reasoning

    A million tokens of context, a step behind Fable on the hardest problems.

    Model ID
    claude-opus-5
    Alias opus
    Context window
    1,000,000tokens
    Also listed as claude-opus-5[1m]

  • Claude Sonnet 5

    Balanced

    The same million-token window, lighter on the clock.

    Model ID
    claude-sonnet-5
    Alias sonnet
    Context window
    1,000,000tokens
    Also listed as claude-sonnet-5[1m]

  • Claude Haiku 4.5

    Fast

    Low latency and high throughput, for work that keeps up with typing.

    Model ID
    claude-haiku-4-5-20251001
    Alias haiku
    Context window
    200,000tokens

Also served

Earlier releases · for pinned clients

1,000,000 tokens of context

  • Claude Fable 5claude-fable-5
  • Claude Opus 4.8claude-opus-4-8
  • Claude Opus 4.7claude-opus-4-7
  • Claude Opus 4.6claude-opus-4-6
  • Claude Sonnet 4.6claude-sonnet-4-6

200,000 tokens of context

  • Claude Opus 4.5claude-opus-4-5
  • Claude Sonnet 4.5claude-sonnet-4-5-20250929
  • Claude Opus 4.1claude-opus-4-1-20250805
  • Claude Opus 4claude-opus-4-20250514
  • Claude Sonnet 4claude-sonnet-4-20250514

GET /v1/models lists all 14, plus a [1m] twin of each 1M-context model so Claude Code can offer the full window. No key needed.

02. Specification

What every key gets.

Nothing to unlock and no tiers of features: all of it works from a key's first request.

Context window
Up to 1,000,000 tokens
Fable 5.1, Opus 5 and Sonnet 5 take a million tokens; Haiku 4.5 takes 200,000. Claude Code's [1m] suffix is accepted as-is.
Usage window
5 hours, rolling, per key
Opens with a key's first request and resets itself five hours later. Usage reads out as a percentage of the window, split by model family:
  • Fable
  • Opus
  • Sonnet
  • Haiku
Streaming
Server-sent events
Set "stream": true and events arrive as the model writes them, with no buffering. Disconnect early and the request is cancelled with you.
Prompt caching
cache_control
Mark a long, stable prefix — a system prompt, a tool schema, a document — and it is cached. Cache reads and writes never count against the window; only fresh input and output do.
OpenAI-compatible
POST /v1/chat/completions
For clients that only speak OpenAI. Requests and responses are translated both ways, streaming and tool calls included — use https://api.claudeopus.pro/v1 as the base URL.
Server-side tools
Web search · image understanding
Both run on our side: nothing to install, no MCP server, no client plugin. Images in a message just work, and each tool is also callable directly under /tools.
Key status
Public, per key
Anyone holding a key can read its status, 5-hour window and reset time on the key status page — no account needed.
Drop-in clients
Claude Code · Cursor · SDKs
The Anthropic SDKs, Claude Code, Cursor, Windsurf, Cline and Roo Code. Change the base URL and the key — nothing else.
Rate limits & expiry
Set per key
Every key carries its own per-minute request ceiling and its own expiry date.
Key issuing
Admins and resellers
Administrators issue keys; resellers issue keys to their own users, from their own quota.

03. Quick start

Two settings. Nothing else changes.

Every client needs the same two facts: the base URL and your key. Same request body, same response, same stream.

  1. Get a key

    Keys start with sk-ant-co- and are issued by your administrator or reseller.

  2. Point your client at the proxy

    Base URL https://api.claudeopus.pro. Routes sit at the root — it's /v1/messages, with no /api in front.

  3. Send the key

    As x-api-key, the way the Anthropic SDKs do, or as Authorization: Bearer — whichever your client already sends. On Claude Code, the one-line setup script writes both for you.

  4. Watch the window

    A key's 5-hour window and reset time are on the key status page, per model.

Environment variables
export ANTHROPIC_BASE_URL=https://api.claudeopus.pro
export ANTHROPIC_AUTH_TOKEN=sk-ant-co-YOUR_KEY

One URL in front of every Claude model.

The docs cover every request and response shape, the limits and the errors. The status page shows what is up right now.

Read the docsService status