fastsecureprivatestabletransparent
real fake intelligence.
An OpenAI-compatible inference API with zero data retention for prompts/completions (see privacy policy). Plug it into OpenRouter (BYOK — coming soon!) or call it directly.
how it works
openai-compatible
Point any OpenAI SDK at RFI by changing the base URL. Same request and response shapes, streaming included — no new client to learn.
prepaid credits
Top up what you want to spend. No subscription, no monthly minimum, and no invoice arriving after the fact.
transparent pricing
One published rate per token type. The numbers below are the same ones the gateway bills against.
openai>=1.0python
import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.rfi.dev/v1",
api_key=os.environ["RFI_API_KEY"],
)
resp = client.chat.completions.create(
model="qwen3.8-27b",
messages=[{"role": "user", "content": "hello"}],
reasoning_effort="medium", # low | medium | xhigh
)pricing · per million tokens
| model | input | cached input | output |
|---|---|---|---|
| qwen3.8-27b | $0.40 | $0.05 | $3.20 |
| more coming soon |
qwen3.8-27b scores 52 on the Artificial Analysis Intelligence Index — the same score as Claude Opus 5 (low), at about a tenth of the price.*
* Per-token pricing; all rates are per million tokens. The comparison is based on the Artificial Analysis Intelligence Index and uses Artificial Analysis's own blended rate (7:2:1 cached input : input : output): $0.44 for qwen3.8-27b at the rates above, against $3.85 for Claude Opus 5 (low) — a factor of 8.8. Scores are Artificial Analysis's, read 2026-08-21, and both models are measured by them rather than by us. Note that qwen3.8-27b reaches 52 at its xhigh reasoning effort while Claude Opus 5 reaches 52 at its low setting; Opus 5 scores higher at its own upper settings.