One base URL in front of the Anthropic API. It routes, meters, and caches every request — then gets out of the way. Your SDK keeps its shape; your keys keep their limits.
▊▊▊▊▊▊▊0M
Tokens of context
Opus 4.8 · Sonnet 4.6
0K
Max output
in one Opus response
0
Model IDs
one base URL
0h
Token window
rolling, per key
No feature flags, no tiers to unlock. It all works the moment a key is issued.
SSE relayed byte-for-byte, with no latency of our own. The stream you get is the stream Anthropic sent.
Cached tokens are free — they never count against the window.
A five-hour token window per key that resets itself.
Per-key request ceilings and expiry dates, enforced upstream.
Admins mint keys; resellers mint keys for their own users, against their own quota. Nobody sees the tier above them.
Web search and image analysis, run on the proxy.
Claude Code, Cursor, Windsurf, Cline — no SDK fork.
A /v1/chat/completions endpoint for the clients that need it.
A million tokens of context on the top two, up to 128K of output in one response — all on the same URL.
claude-opus-4-8The flagship — a million tokens of context and a 128K ceiling on a single answer.
context
1,000,000
max output
128,000
claude-sonnet-4-6The same million-token window, lighter on the clock. The default for most prompts.
context
1,000,000
max output
64,000
claude-haiku-4-5Low latency, high throughput — for work that keeps up with typing.
context
200,000
max output
8,192
Point the base URL at the proxy and send your key. Same request body, same response shape, same stream — on every model.
▊▊▊▊▊▊▊▊▊▊▊▊▊▊Budgets, windows, and multi-tenant keys — for teams who ship.