SignedEvery response is cryptographically verifiable

The GPT-5.5 API,25% cheaper.

The exact model OpenAI serves, running on Azure. Point your client at our base URL, keep the rest of your code, and pay less on every token.

Works with

OpenAI SDKVercel AI SDKLangChainLiteLLMCursor
quickstart.py
python
from openai import OpenAI
client = OpenAI(
base_url="https://www.draftworks.dev/v1",
api_key="dw_live_...",
)
# nothing else changes
25%
below OpenAI list
~450ms
to first token
1M
token context window
100%
OpenAI-compatible
Why Draftworks

Cheaper, quick, and you can prove what ran.

Served from Azure

Same weights as api.openai.com, with less sitting in front of them. First token usually lands under half a second.

draftworks~450ms
openai direct~780ms

Typical streaming time-to-first-token from us-east. Don't trust the bar: every response ships its own timing in usage.latency_checkpoint.

Up to 25% cheaper

Every token costs less than OpenAI list. You start at 20% off and drop to 25% once you've bought $5,000 in credits. One credit is one dollar of inference.

Input$4.00$5.00
Output$24.00$30.00
Cached input$0.40$0.50

Per 1M tokens.

Signed on every response

Some resellers quietly route you to a cheaper model. We can't. Each response is signed with Ed25519 over the model id and a hash of the exact bytes we sent back.

verified
model gpt-5.5-2026-04-24
signature ✓ valid

Check any response yourself, no account needed.

Verification

Proof you got GPT-5.5, not a knockoff.

When a reseller sits between you and the model, the response alone rarely tells you what actually ran. So we sign it. Every reply carries an Ed25519 signature over the model id, the token counts you paid for, and SHA-256 hashes of your exact bytes in and out.

Pull any receipt from GET /v1/receipts/{id} and verify it against our public key, in your own code or straight from the browser.

x-draftworks-receipt
ed25519
{
"v": 1,
"id": "chatcmpl-Dz46ISfIOZxiNscV…",
"endpoint": "chat.completions",
"model": "gpt-5.5-2026-04-24", // the proof
"stream": false,
"prompt_sha256": "9d2f…c41a", // your bytes
"response_sha256": "41c8…0be7", // our bytes
"input_tokens": 1204,
"cached_input_tokens": 512,
"output_tokens": 386,
"total_tokens": 1590
}
Pricing

Priced per token, always under list.

You buy credits up front. One credit is one dollar of inference at your rate. No subscription and no seat fees. Rates below are per 1M tokens, next to OpenAI list.

Base

−20%

From your very first dollar.

Input$4.00$5.00
Cached input$0.40$0.50
Output$24.00$30.00

$5 minimum to start.

Scale

−25%

Once lifetime purchases reach $5,000.

Input$3.75$5.00
Cached input$0.375$0.50
Output$22.50$30.00

Applied automatically, forever.

Reasoning tokens billed as output, standard across the industry.

Full pricing & calculator

Your prompts stay yours

Nothing you send trains a model. We keep SHA-256 hashes and token counts for billing and receipts, and the Responses API store flag defaults to off.

Privacy policy

One line to switch. 25% off to stay.

Create a key, add $5 in credits, and point your existing OpenAI code at a new base URL.