Pay once for repeated work
Caching
Serve identical requests from the gateway at $0, and use provider prompt caching to discount long, reused prompts.
usage · cached response
{"usage": {"prompt_tokens": 12,"completion_tokens": 48,"cost": 0},"metadata": { "cached": true }}
Repeated identical requests cost $0
01
$0 cache hits
Identical requests never reach the provider.
02
Prompt caching
Discounted reuse of long prefixes.
03
Cache modes
Automatic, client-managed or off, 5m or 1h TTL.
04
Visible savings
Cached tokens shown in every response.
Built for
Coding tools
Huge system prompts, reused every call
Test suites
Re-run evals without paying twice
FAQ bots
Common questions served instantly
Mark a prefix for caching
request.json
1{2 "model": "claude-sonnet-4-6",3 "messages": [4 {5 "role": "system",6 "content": [7 {8 "type": "text",9 "text": "<long system prompt>",10 "cache_control": { "type": "ephemeral" }11 }12 ]13 },14 { "role": "user", "content": "Review this diff" }15 ]16}