What interwebs.ai is — and how to use it
interwebs.ai is a unified gateway for AI compute. The same frontier models you already use, behind one OpenAI-compatible API and one wallet, at the real provider price plus a flat 5% — with a 50%-off batch lane for anything that can wait an hour. Under the hood, a live price-matching engine routes every request to the cheapest model that clears your quality bar. Every price on the board is a real provider rate; nothing is simulated; every response carries a receipt.
What it does for you
One endpoint, every model
Text, image, video, PDF and audio behind a single OpenAI-compatible API and a single prepaid wallet. Swap between GPT, Claude, DeepSeek, GLM, FLUX, Veo, Sora and more by changing one string.
The only fee is +5%
Every request bills at the raw provider price plus a flat 5%. No subscription, no seats, no minimums — and the receipt on every response shows raw vs charged, to the nano-dollar.
Batch lane: −50%
Non-urgent text work runs async (usually under an hour, ≤24h worst case) on the provider’s own 50%-off Batch API. Half the list price, plus the same 5%. A real provider discount, passed through.
Price-aware routing
Ask for a quality tier instead of a model and each request routes to the cheapest live model that clears the bar — re-priced from the live board on every call, escalating automatically if a provider fails.
Custom routers
Build your own model id: categories in plain English, each routing to a pinned model or a live-repriced tier. Call it as model:"router:<your-slug>" from any OpenAI SDK.
Built for agents
Scoped revocable keys, per-key spend mandates, auto-reload, and an idempotent append-only ledger — so an autonomous agent can spend unattended without ever exceeding the budget you set.
Use it from your own AI in three steps
- Sign up and add funds. Create an account, then top up by card on the Overview tab (Stripe Checkout — unused balance is refundable any time). You only ever spend as requests bill.
- Mint an API key. Dashboard → API → Create key. It is shown once; keys are scoped and revocable in one click.
- Point your existing OpenAI SDK at the gateway. Change the base URL, nothing else:
from openai import OpenAI
client = OpenAI(
base_url="https://gateway.interwebs.ai/api/v1",
api_key="iwk_live_…", # minted in your dashboard
)
r = client.chat.completions.create(
model="auto", # cheapest live model that clears standard quality
messages=[{"role": "user", "content": "Summarise this contract…"}],
)
print(r.choices[0].message.content)
print(r.model_extra["interwebs"]) # the receipt: raw cost, your +5%, routing trailThe model string is the whole interface: auto (cheapest live model at standard quality), auto:economy / auto:standard / auto:premium (set the quality bar), any exact model id from the spot board (e.g. claude-sonnet-4-6), or one of your own custom routers:
curl https://gateway.interwebs.ai/api/v1/chat/completions \
-H "Authorization: Bearer iwk_live_…" \
-H "Content-Type: application/json" \
-d '{"model":"router:support-agent",
"messages":[{"role":"user","content":"I was double-charged…"}]}'Build a custom router in two minutes
A router is your own model id. Instead of hard-coding which model handles what, you describe your request types in plain English and each call routes itself:
- Dashboard → API → New router. Name it — the slug becomes your model id, e.g. router:support-agent.
- Add categories: a name (“Coding”), a when-to-use description (“programming, debugging, code review”), and a target — a pinned model, or a price tier that re-routes to the cheapest live model on every single call.
- Set the “everything else” fallback, optionally a system prompt and sticky routing (one model per conversation — best for coding agents), and save.
- Call it with model:"router:<slug>". One cheap classification call (≈$0.0001, itemized on your receipt) picks the category; the target model answers.
Onboard users from your own tool — and earn on their spend
Your app or agent can create accounts for its users machine-to-machine — no browser, one call. Pass your referral code and the new account is attributed to you permanently: you earn 2% of everything it ever spends (and 1% from anyone it refers), carved out of our 5% fee. The response mints the account's first API key; its owner funds the wallet and can claim the dashboard with the same email and password any time.
curl https://gateway.interwebs.ai/api/v1/signup \
-H "Content-Type: application/json" \
-d '{"email":"new-user@example.com",
"password":"…",
"ref":"YOURCODE",
"key_name":"made-by-your-tool"}'
# → 201 {
# "user_id": "…",
# "referred": true,
# "api_key": { "token": "iwk_live_…" } # shown once — hand it to your user
# }Attribution is first-touch and set only on brand-new accounts — an existing account can never be re-attributed, so nothing downstream of you can hijack your referrals. Invalid codes never block a signup; they just don't attribute. Signups are rate-limited per IP, and there are no per-head bounties: commission accrues only on real spend.
Live prices — raw cost + 5%
Every figure below is what a request actually bills — the real provider rate plus our flat 5%, with the raw cost printed underneath so you can check the math. These tables render from the same price table the router charges from, so they can never disagree with your receipt.
Free models — $0.00, unlimited
| Model | Provider | You pay |
|---|---|---|
| glm-4.7-flash | zhipu | $0.00 |
| glm-4.5-flash | zhipu | $0.00 |
These run on the AI labs’ own free tiers — genuinely $0.00, no daily cap from us, though the labs’ rate limits apply and they may use prompts for training. Paid models never share your data.
Chat & text models — per 1M tokens
| Model | Provider | Input | Output | Batch output |
|---|---|---|---|---|
| deepseek-v4-flash | deepseek | $0.147 cost $0.14 | $0.294 cost $0.28 | — |
| gpt-4o-mini | openai | $0.1575 cost $0.15 | $0.63 cost $0.60 | $0.315 |
| deepseek-v4-pro | deepseek | $0.45675 cost $0.435 | $0.9135 cost $0.87 | — |
| glm-4.6 | zhipu | $0.63 cost $0.60 | $2.31 cost $2.20 | — |
| glm-5.2 | zhipu | $1.47 cost $1.40 | $4.62 cost $4.40 | — |
| claude-haiku-4-5 | anthropic | $1.05 cost $1.00 | $5.25 cost $5.00 | $2.625 |
| claude-sonnet-4-6 | anthropic | $3.15 cost $3.00 | $15.75 cost $15.00 | $7.875 |
| claude-opus-4-8 | anthropic | $5.25 cost $5.00 | $26.25 cost $25.00 | $13.125 |
The same rates everywhere a model answers — the assistant, the API, tools. Batch output is the provider’s own 50%-off async lane (usually under an hour, ≤24h worst case) plus the same 5%, available through the API.
Image generation
| Model | Provider | You pay | Unit |
|---|---|---|---|
| flux-schnell | bfl | $0.00315 cost $0.003 | per image |
| gpt-image-2 | aivideoapi | $0.024675 cost $0.0235 | per image |
| flux-dev | black-forest-labs | $0.02625 cost $0.025 | per image |
| ideogram-v2a-turbo | ideogram-ai | $0.02625 cost $0.025 | per image |
| seedream-3 | bytedance | $0.0315 cost $0.03 | per image |
| flux-1.1-pro | bfl | $0.042 cost $0.04 | per image |
| recraft-v3 | recraft-ai | $0.042 cost $0.04 | per image |
| nano-banana-2 | aivideoapi | $0.084 cost $0.08 | per image |
| ideogram-v4 | aivideoapi | $0.147 cost $0.14 | per image |
Video generation
| Model | Provider | You pay | Unit |
|---|---|---|---|
| seedance-1-pro | bytedance | $0.0315 cost $0.03 | per second |
| veo-3.1 | aivideoapi | $0.042 cost $0.04 | per second |
| kling-v1 | kling | $0.0525 cost $0.05 | per second |
| wan-2.2-t2v-fast | wan-video | $0.0525 cost $0.05 | per video |
| kling-v1.6-standard | kwaivgi | $0.0525 cost $0.05 | per second |
| wan-2.7 | aivideoapi | $0.126 cost $0.12 | per second |
| sora-2 | aivideoapi | $0.13125 cost $0.125 | per second |
| kling-3.0 | aivideoapi | $0.147 cost $0.14 | per second |
| seedance-2.0-fast | aivideoapi | $0.160125 cost $0.1525 | per second |
| seedance-2.0 | aivideoapi | $0.1995 cost $0.19 | per second |
| kling-v2.1-master | kling-pro | $0.294 cost $0.28 | per second |
| kling-v2.0 | kwaivgi | $0.294 cost $0.28 | per second |
| video-01 | minimax | $0.525 cost $0.50 | per video |
Video bills per second of generated footage at the rate shown.
Talking with your assistant
| Direction | You pay |
|---|---|
| Your voice heard (speech → text) | ~$0.00315 / min cost ~$0.003 / min |
| Replies spoken back (text → speech) | ~$0.01575 / min cost ~$0.015 / min |
Voice is metered on the provider’s exact usage tokens whenever they’re returned — the per-minute figures are the approximate equivalent. A typical spoken exchange costs a few tenths of a cent, itemized under the reply.
Tools
| Web search | $0.00 |
Anything else — image jobs, deep research — bills the same way: the metered provider cost plus 5%, itemized on your receipt.
Common questions
Is it actually cheaper than going direct?
How do I know the prices are real?
What happens if a provider fails mid-request?
Do I need a subscription?
How does refer & earn work?
Can my agent top itself up?
Try it with a dollar
A dollar of balance is thousands of routed requests.
Create your free account