Inference at the
market price,
never the list price.
Drift is an order book for model credits. Resting balances fill your order deepest-discount first, so every request through one key costs less than the lab charges — settled per call.
- Depth
- $21500
- Best bid
- 30%
- Models
- 12
Order book
- L1Protocol$120030%
- L2Protocol$240025%
- L3Protocol$380020%
- L4Protocol$260018%
- L5Protocol$420015%
- L6Protocol$300012%
$100 of credits costs
$70.00
/api/v1/chat/completionsOpenAI15 tokens metered · $0 saved vs list
- Face on the book
- $21500
- Listed and ready to fill
- Deepest discount
- 30%
- Fills first, FIFO per level
- Models on one key
- 12
- OpenAI-compatible relay
- Saved vs list
- $0.00
- Across metered requests
Market / Buy
Your inference, at a discount.
Credits are dollars of inference. One credit is $1 of usage.
Credits
$100
Discount applied
0%
You pay
$100.00
You save $0.00 on this order
0 fills
No subscription. Paid from your balance, settled instantly.
Live order book
Every level is real credit, resting until it fills.
Sellers escrow balance at a discount they choose. Your order walks the book from the deepest discount down, and each fill settles to the seller instantly, minus the protocol fee.
- Depth
- $21500
- Best
- 30%
- Average
- 16.6%
Depth, best first
8 levels · $21500 face
- L1Protocol$120030%
- L2Protocol$240025%
- L3Protocol$380020%
- L4Protocol$260018%
- L5Protocol$420015%
- L6Protocol$300012%
- L7Protocol$250010%
- L8Protocol$18008%
Live from the ledger
Every request is counted, priced and settled.
These figures come straight from the usage ledger — the same records that debit your balance and appear on your account page.
- Tokens processed
- 15
- 15 tokens metered through the gateway
- Saved vs list price
- $0.00
- What those same tokens would have cost at the providers' list rates
Every model, one key
12 models, below what the lab charges.
Prices are per one million tokens. The market price is what you actually pay on your key.
GPT-5 Mini
openai/gpt-5-mini
context 400K
$0.25 / $2
$0.20 / $1.56
−22%
Gemini 3 Flash
google/gemini-3-flash
context 1M
$0.30 / $2.50
$0.23 / $1.88
−25%
GPT-5.6 Luna
openai/gpt-5-6-luna
context 400K
$1 / $6
$0.80 / $4.80
−20%
GPT-5.4
openai/gpt-5-4
context 400K
$1.25 / $10
$1.02 / $8.20
−18%
Gemini 3.1 Pro
google/gemini-3-1-pro
context 2M
$1.25 / $10
$1.02 / $8.20
−18%
GPT-5.6 Sol
openai/gpt-5-6-sol
context 1M
$3 / $15
$2.55 / $12.75
−15%
Claude Sonnet 4.6
anthropic/claude-sonnet-4-6
context 200K
$3 / $15
$2.55 / $12.75
−15%
Claude Sonnet 5
anthropic/claude-sonnet-5
context 1M
$3 / $15
$2.49 / $12.45
−17%
Claude Opus 4.6
anthropic/claude-opus-4-6
context 200K
$5 / $25
$4.30 / $21.50
−14%
Claude Opus 4.7
anthropic/claude-opus-4-7
context 200K
$5 / $25
$4.35 / $21.75
−13%
Claude Opus 4.8
anthropic/claude-opus-4-8
context 200K
$5 / $25
$4.40 / $22
−12%
Claude Opus 5
anthropic/claude-opus-5
context 200K
$15 / $75
$13.20 / $66
−12%
Buy, key, request
Three steps from balance to a cheaper bill.
01
Buy credits
Pick an amount. Your order fills the deepest discount on the book first, so you always get the best price available at that moment.
02
Create one key
Your credits become API balance behind a single key that reaches every model on the catalogue.
03
Send requests
Point your OpenAI or Anthropic client at the gateway. Every request is metered at that model's discounted rate.
Migration
two lines change: base URL and key
curl https://your-app.base44.app/api/v1/chat/completions \
-H "Authorization: Bearer $SPREAD_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "anthropic/claude-sonnet-4-6",
"messages": [{ "role": "user", "content": "Hello" }]
}'Same models, less spend
What changes is the price, never the answer.
The gateway is a relay. Your request goes to the model you named, unchanged, and the response comes back carrying the provider's own model id.
List price
$2000000
With Drift
$1560000
You save $440000
22% less
openai/gpt-5-mini · $2.00 per 1M output tokens · 1M tokens
Migration
The whole migration is two lines.
Change the base URL and the key. Model names, streaming, tool calls and vision behave exactly as they do today.
Before
OPENAI_BASE_URL=https://api.openai.com/v1 OPENAI_API_KEY=sk-…
With your key
OPENAI_BASE_URL=https://<your-app>/api/v1 OPENAI_API_KEY=sk-srp-…
Replace <your-app> with the domain this app is published on, then use any key you create on your account page.
One market, clear rules
Built around honest inference.
Pay less for the same model
Buy credits below face value from balance that someone else isn't using. One key reaches every model at its discounted rate.
Sell what you don't use
List unused credits at a discount level. Buyers fill the deepest level first, oldest listing first within it.
Turn balance into more balance
Buy a discounted block, keep what you need, re-list the rest. The book is shared, so your listing is visible to every buyer.
Relay, never rewrite
Requests go to the model you named. Nothing is substituted, downgraded or quietly routed to a smaller model.
Keep prompts private
We store the model, token counts, cost and timing for billing. The prompt text and the answer are never written to our records.
Settle per request
Every request debits your balance the moment it returns, and appears in your usage ledger with its own cost and latency.
Questions
Everything about credits.
The mechanics that decide your bill, answered without the marketing.
A fixed amount of API credit. One credit is one dollar of model usage, so $100 of credits is $100 of inference through your Drift key, spent at each model's discounted rate.
From balance other accounts aren't using. Sellers list credits below face value; buyers fill the deepest discount first, oldest listing first at the same level. Anything left over fills at the protocol floor.
Yes. The gateway is a relay: your request goes to the model you named, and the response carries that model's own name and id. Nothing is substituted or downgraded.
No. We keep billing metadata only: the model, token counts, cost, latency and status of each request. Prompts and responses pass through and are not written to our records.
A managed upstream account behind the gateway. You authenticate with your own Drift key and never hold a provider key of your own.
The base URL and the key. Model names, streaming flags and message shapes stay as they are — the endpoint is OpenAI-compatible.
No subscription. The smallest order is $5 of credits, and you can start with the free trial credit on a new account.
Credit is a grant of product access and is not redeemable for cash. If something goes wrong with a purchase, it stays visible in your ledger so it can be corrected.
Start on the book
Every model, one key.
You keep the spread.
New accounts start with trial credit. Buy only what you burn, at the price the book is quoting right now.