Three flat rates. No subscription tiers. No minimums. No per-seat charges. You pay markup on top of provider cost — and that's the whole story.
Pick a mode + model + monthly token volume. The numbers below are all live; we maintain the price book at docs/pricing/price-book.json.
Provider list prices in USD. Local models bill against a cloud-equivalent basis at 5% markup. Move the slider to see how the typical-chat blended rate moves with your input/output ratio.
| Model | Input / 1M | Output / 1M | Typical chat / 1M |
|---|---|---|---|
| Anthropic · Claude Opus 4.8 | $5.00 | $25.00 | — |
| Anthropic · Claude Sonnet 5 | $3.00 | $15.00 | — |
| OpenAI · GPT-5.5 | $5.00 | $30.00 | — |
| OpenAI · GPT-5.4 | $2.50 | $15.00 | — |
| xAI · Grok 4.3 | $1.25 | $2.50 | — |
| Google · Gemini 3.5 Flash | $1.50 | $9.00 | — |
| DeepSeek · V4 Flash | $0.14 | $0.28 | — |
| Local · Llama 3.2 8B (Ollama, 16+ GB VRAM) | $0.00 (own GPU) | $0.00 (own GPU) | — |
| Local · Llama 3 70B (Ollama, 80 GB VRAM) | $0.00 (own GPU) | $0.00 (own GPU) | — |
| Local · Qwen 2.5 7B (Ollama, 16+ GB VRAM) | $0.00 (own GPU) | $0.00 (own GPU) | — |
Prices in cents per 1M tokens stored as data-in / data-out. "Typical chat" = blended rate at the selected ratio (default 1 : 4 — 1 input token per 4 output tokens — matching most assistant-style traffic). Local rows show the 5% CloakAPI markup against a cloud-equivalent basis, since the model itself runs free on your hardware.
No tier comparison columns. There aren't any. Every account gets all of these.
Every privacy-preserving AI gateway promises to handle your data carefully. CloakAPI promises something stronger. On the client-side ways — desktop app, browser extension, local proxy, and the chat, which tokenises in your own browser — we never see your plaintext PII at all: tokenisation runs on your device before any byte leaves it, so the math says we can't read your prompts even if we wanted to, and that fact is itself attested in every receipt. Our gateway is a blind token relay — it forwards only the tokens your device produced and never inspects, detects, or tokenises content on our servers. It never tokenises on our servers, so to protect a plain SDK you point it at the local proxy and tokenisation stays on your own machine. We never hold your plaintext — which is why many customers conclude that, on those paths, we aren't a Business Associate, Processor, or Service Provider the way a gateway that stores plaintext is (a structural argument to evaluate with your own counsel — not legal advice).
Every reply ships with a chain-linked, ECDSA-P-256-signed receipt envelope. Any auditor can verify it offline against our public JWKS at signedreceipts.org/verifier — no CloakAPI account, no cooperation from us required. We don't know of a competing gateway that publishes this artifact today. It's the thing that ends "is this provider trustworthy" conversations, because the answer stops depending on trust.
Our markup is structural — we have to make money per call — but it's flat: 15% on pooled keys, 5% on BYOK, 5% on local. There is no "Enterprise" tier that unlocks SSO, audit logs, BAA, or compliance packs. Every API key gets every feature, on day one, at the same rate.
docs/pricing/price-book.json is source-of-truth and gets a new version every time a provider posts a change; subscribe to its RSS for advance notice.