Skip to content
OP Inference
Multi-modal inference gateway

One API.
Every model.
Never down.

Text, image, video, audio and music from every major provider through one OpenAI-compatible endpoint. The best price and speed on every call, automatic failover before the first token, and EU or global routing set per key.

Get an API keyBrowse modelsRead the docs

Providers reachable through OP Inference: OpenAI, Anthropic, Google, Meta, Mistral AI, DeepSeek, Qwen, Moonshot AI, Z.ai, Black Forest Labs, ElevenLabs, Stability AI.

// The gateway

Above every provider. The best route for every call.

Five modalities, one key

Text, image, video, audio and music sit behind the same base URL and the same key. Ask for a model by its published id and that is the model that answers, never a substitute under the name you asked for.

Region-pinned, enforced per key

Pin a key to EU routes and inference stays on them, or leave it open to route across every available provider. No-training policies, zero data retention and a retention class on every request, bought as a policy rather than as a contract negotiation.

Failover before the first token

A request that fails before it has sent anything moves to another endpoint serving the same model. One provider having a bad night stops being your bad night, and every generation still leaves exactly one durable usage record.

Drop-in

Change one line of code.

We are OpenAI-compatible, so any SDK or agent framework that speaks OpenAI works as is. Point it at our base URL, keep your streaming code, and switch models by changing a model name instead of rewriting an integration.

  • Python, Node, Go, Rust, or anything else with an OpenAI client
  • Routing policy, budget and expiry set per key
  • Add sign=true to get a Content Passport on generated media
client.py
from openai import OpenAI

# Your existing client, your existing streaming code,
# your existing agent framework. All of it keeps working.

client = OpenAI(
One API, every modality

Five endpoints.
One key behind all of them.

Most teams end up with a key, an invoice and an integration per provider, and a rewrite every time a better model appears. Here every modality answers on the same base URL, under the same key, with the same spend limit.

  • Text

    POST /v1/chat/completions

    Streaming, tool calls, structured output, reasoning

  • Image

    POST /v1/images/generations

    Generation and editing

  • Video

    POST /v1/videos

    Text-to-video and image-to-video, as a job you poll

  • Audio

    POST /v1/audio/speech

    Speech synthesis and transcription

  • Music

    POST /v1/audio/music

    Generation from a prompt

  • Same key

    One invoice, prepaid credits, a spend limit and an expiry per key, across every endpoint on this list.

European by default

Residency is a setting on your key.

  • Zero data retention

    We do not store prompts or outputs, and we do not log them as content. Every request carries a retention class, every sink honours it, and an hourly purge proves it.

  • EU or global routing

    Pin a key to EU routes and inference stays on them, or leave it global to route across every available provider. You set it per key in the dashboard.

  • Provenance on request

    Add sign=true and generated media comes back with a C2PA Content Passport and a watermark, signed in the same call that made it.

  • GDPR aligned
  • EU or global routing
  • Zero data retention
  • No training on your traffic
  • C2PA provenance
How it works

Every call takes the best route. Every failure has a detour.

  1. 01.

    Request

    Chat, reasoning, image, video, audio, music or tool use, all as one OpenAI-compatible call against one base URL. You need no per-provider SDK and no second set of credentials.

  2. 02.

    Compare

    We measure price, latency, uptime and feature support across every route that can serve the model you asked for. That comparison happens before we dispatch your call, not on a report about calls you already paid for.

  3. 03.

    Route

    Your policy decides: cheapest, fastest, EU-only, global or quality-first. The same step checks the budget and model allowlist on your key, so a runaway agent stops at its own ceiling.

  4. 04.

    Deliver

    The answer streams back. If a route fails before it has sent anything, the request moves to another endpoint serving the same model. We write one usage record per generation, whatever happened upstream.

Software routing, today. Our own European capacity underneath it, next.

Who it is for

Anyone who buys tokens.

Every model in one account.

Ship the best model for each job without ten keys, ten invoices and an integration that breaks the week a better model lands. New models are a string change.

# Swap the model, keep the code.
reply = client.chat.completions.create(
    model="anthropic/claude-opus-5",   # or any id in the catalog
    messages=[{"role": "user", "content": prompt}],
    stream=True,
)
The market

You are buying a commodity with no market maker.

Four numbers explain why a router pays for itself. Every one of them is published by somebody other than us.

  • 8.9×

    Price spread on one model

    The same model costs up to nine times more depending on which route you happen to buy it from, and the spread moves every week.

    Artificial Analysis provider comparison, September 2026

  • 39×

    Speed spread on one model

    Latency across those same providers differs by a factor of thirty-nine. Picking a provider once and never revisiting it is a performance decision you are still paying for.

    Artificial Analysis provider comparison, September 2026

  • 47

    Provider incidents in four months

    Incidents logged on one major provider's status page between June and September 2026, twenty-seven of them in July alone. One provider is one point of failure.

    status.openai.com, retrieved September 2026

  • $1.01

    Blended price per 1M tokens

    Tokens are a commodity with a published daily price and no market maker. Nobody is shopping the spread on your behalf, until you put a router in front of it.

    Silicon Data blended token price index, 19 September 2026

Routing policies

You set the rule. We find the route, call by call.

Pick a policy per key. Automatic retry and failover, per-key budgets and spend alerts are on underneath all five of them, and custom policies are available on committed-use accounts.

  • Cheapest

    The lowest price that still meets your model and feature needs.

  • Fastest

    The lowest latency right now, for live products and agents.

  • EU-only

    Inference stays on EU routes, enforced per key.

  • Global

    Routes across every available provider, for the widest choice.

  • Quality-first

    The strongest model for the task, with a fallback ready.

The idea

Tokens became a utility. Buying them didn’t.

Electricity has a grid that dispatches to the cheapest available plant. Tokens don’t. So we built the dispatcher: one API above every provider that compares price, speed and availability on every call, routes to the best qualifying supply, and fails over the moment one of them goes quiet.

What changes on day one.

Onboarding is an afternoon: a key and starter credits, a base-URL change, then your routing policy and budgets per key.

Without OP Inference

  • A key and an invoice per provider
  • A provider outage takes you down
  • You pay the first price you see
  • Switching models means a rewrite
  • Data may leave the EU
  • AI output leaves unlabelled

With OP Inference

  • One key, one invoice, prepaid credits
  • Automatic failover to the next best route
  • Live price and speed comparison on every call
  • Change a model name, keep your code
  • EU or global routing, set per key
  • sign=true returns a Content Passport
Pricing

Our fee is one line on the invoice.

We add 4.5% on top of the provider’s own token price, and that price is never marked up. Self-serve gateways typically charge 5% to 5.5%. Every model’s live price sits in the dashboard before you spend a cent.

  • Self-serve

    Prepaid credits

    Create an account, mint a key, top it up. Every model shows its live price in the dashboard before you spend anything.

    • Live per-model prices and speed
    • Spend limit, expiry and model allowlist per key
    • Usage analytics and spend alerts
    • All five modalities on one key
    Create an account
  • Best value at volume

    Committed use

    Annual agreement

    Volume discounts on pooled spend, EU or global routing, invoicing and a named contact. Priced to your models and volumes.

    • Enterprise discount on routed spend
    • EU or global routing set per key
    • Invoicing, SLA and dedicated support
    • Custom fallback policies, audit logs, SSO
    Contact sales

  • Resellers

    Credit batches

    Telcos, ISPs, integrators and platforms bundling prepaid AI credits for their own customers, with a margin on top.

    • Prepaid credit batches at a discount
    • Sub-keys with their own budgets
    • Per-project usage reporting
    • White-glove onboarding
    Talk to us

Enterprise discounts, EU-only routing and private endpoints are quoted per account. Write to [email protected] or use the form on the contact page.

Common questions

The questions we
actually get asked.

You keep one key, one invoice and one integration instead of ten. On every call we compare price, latency and availability across every route that can serve the model you asked for, and dispatch to the one your policy prefers. When a route fails before it has sent anything, the request moves to another endpoint serving the same model.

Inside the gateway

Product notes & engineering decisions.

Get started

Live in an afternoon: create a key, top up starter credits, change the base URL, then set your routing policy and budgets per key.

Never pay the first price.
Never go down at night.

Start self-serve with a key and prepaid credits, or book a thirty-minute call and we will scope your models, volumes and routing rules with you.

Get an API keyTalk to salesDocumentation

or write to [email protected]