Credits & Pricing
Proxyify uses a prepaid credit system. Credits are deducted per request based on the model's pricing. They never expire.
Credit packages
| Package | Price | Credits | Notes |
|---|---|---|---|
| Free | $0 | 500 | One-time on sign-up. No card required. |
| Starter | $9 | 12,000 | |
| Pro | $29 | 30,000 | |
| Scale | $99 | 120,000 |
Credits can be topped up at any time from Dashboard → Billing. Multiple packages can be purchased — credits stack.
How credits work
- Credits are deducted only on a successful provider response (HTTP 200). Errors are free.
- Credits never expire and never reset monthly.
- Every response includes a
_proxyifyobject showing exact usage:
{
/* ...provider response fields (id, choices, usage, etc.) ... */
"_proxyify": {
"credits_used": 24,
"cost_usd": 0.024,
"model_used": "openai/gpt-4o",
"latency_ms": 1240,
"cached": false,
"cheaper_model": "openai/gpt-4o-mini" // cheaper alternative with same capabilities, or null
}
}
Pricing by modality
Text (Chat)
Billed per token. _proxyify includes input_tokens and output_tokens.
Image generation
Billed per image. Cost depends on model and resolution. _proxyify includes images_count.
Video generation
Billed per second of generated video. _proxyify includes duration_seconds. Credits are charged when the job completes (polling response).
Speech-to-Text (STT)
Billing varies by model:
- Whisper models — billed per audio second (
audio_seconds) - GPT-4o Transcribe, Chirp 3 — billed per token (
input_tokens/output_tokens)
Text-to-Speech (TTS)
Billed per character of input text. Credits appear in the X-Proxyify-Credits-Used response header (TTS responses are raw audio streams, not JSON).
Partial billing
Streaming introduces edge cases:
| Scenario | Behaviour |
|---|---|
| Error before any tokens sent | No credits deducted. Standard JSON error response. |
| Error mid-stream | Credits deducted for tokens already generated. Status stays 200. Stream ends with finish_reason: "error". |
| Client disconnects (cancels) | For providers that support cancellation (OpenAI, Anthropic) — charged for tokens generated so far. For providers that do not (Google, AWS Bedrock) — full response is generated and billed. |