Every model.
One key.
Paid in crypto.

431 LLMs, image, video and editing models for your agents. One API, one MCP server, paid in USDG or ETH.

curl https://omnirail.org/api/v1/images \
  -H "Authorization: Bearer $OMNIRAIL_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2",
    "prompt": "Editorial product photograph of a polished brass espresso machine on a travertine counter, morning window light",
    "aspect_ratio": "4:5"
  }'

Editorial product photograph of a polished brass espresso machine on a travertine counter, morning window light

openai/gpt-image-2 · from $0.077

With OmniRail you can ↴

Real outputs, real prices.

Every image below came out of openai/gpt-image-2, one of 49 image models behind the same key.

Explore all models

Tiny isometric city built from server racks and glowing cables, pastel clay render
openai/gpt-image-2

Portrait of an elderly fisherman in a yellow oilskin on a grey Brittany harbour, film grain
openai/gpt-image-2

Risograph print of a robotic hand holding a glowing coin, cobalt and tangerine inks
openai/gpt-image-2

Macro photograph of a hummingbird drinking from a vivid orange flower
openai/gpt-image-2

Brutalist concrete house cantilevered over a sea cliff at blue hour, long exposure
openai/gpt-image-2

Provider price + 15%.

No subscription, no minimum, credit that never expires. Every call reserves its worst case and pays what the provider billed. Failed calls are free.

Language · per 1M tokens in / out

Images · per image

Video · per second

Estimate a call

ModelInput tokens · your prompt, files and history2,000Output tokens · the answer, reasoning included500

Prices come from the provider in real time. You are charged what the provider reports for the call, not this estimate.

ESTIMATE · 1 CALL

anthropic/claude-opus-5.5

2,000 in · 500 out

Provider cost$0.0180

Markup 15%$0.00270

$RAIL promo (when funded)−$0.00135

TOTAL$0.0207

× 1,000 calls$20.70

One message. Your agent installs itself.

Paste it into Claude Code, Codex, Cursor or any agent. It reads the guide, adds the MCP server and asks you for a key or a wallet.

Read https://omnirail.org/agents.md and set yourself up with OmniRail so you can use any AI model (LLMs, images, video, edits).

Paste into Claude Code, Codex, Cursor, OpenClaw, Hermes… The agent reads the guide, installs the MCP and asks you for a key or a wallet.

Crypto is the only checkout.

You decide who holds the wallet.

01 · you

Top up from your wallet

Sign in with a signature, send USDG or ETH on Robinhood Chain, and give each agent its own key with a spend limit. Revoke any key in one click.

Open the dashboard

02 · your agent

Let it pay per call

Any endpoint answers 402 with a USDG price. The agent signs, we settle on-chain and run the call. No account, no card, no gas.

→ POST /v1/images
← 402  0.089 USDG
→ PAYMENT-SIGNATURE (EIP-3009)
← 200  PAYMENT-RESPONSE tx 0x…

$RAIL · on Pons

A token that makes every call cheaper.

A share of $RAIL's trading fees refills a public pool on Robinhood Chain. While it has funds, everyone's markup is cut in half, automatically. No holding requirement.

See the pool

Now: 15%
Funded: 7.5%
Pool: $0

Questions

Which models?: Every model OpenRouter serves with a public price, across text, images, edits and video. The catalog updates live.
What do I pay with?: USDG or ETH on Robinhood Chain. x402 uses USDG with gasless authorizations; agents never need ETH.
Can an agent run up a bill?: Keys carry spend limits, every call reserves its maximum first, and the local MCP enforces a daily budget.
Do you keep my prompts?: No. We keep billing records only: model, cost, time.
Do I need the token?: No. $RAIL only feeds the pool that lowers everyone's markup.
What if a call fails?: Provider errors, filtered images and failed videos cost nothing.

Every model.One key.Paid in crypto.