Open market for AI inference

Buy inference at the market floor. Sell your spare capacity.

One OpenAI-compatible endpoint. Every provider competes to serve each request and you pay the winning price. Want a specific one? Pin it. Got spare capacity? List it and earn. Sellers keep 95%.

The problem

One channel is
expensive and exposed.

Betting everything on a single inference provider means you overpay, you can't move, and one outage takes your product down with no way to reroute.

01

Inference you overpay for

List prices swing week to week and you have no leverage. Bound to one channel, you quietly overpay on every single token.

02

Vendor lock-in

One SDK, one bill, one rate limit. Switching channels means a rewrite, so you stay put even when a cheaper, faster route opens up.

03

Downtime with no second route

When your one channel degrades or rate-limits you, your product degrades with it. There is no other route to reroute to.

04

A fragmented ecosystem

Every provider ships a different API, pricing model, and latency profile. Comparing them is a spreadsheet, not a decision.

The solution

One current across every provider

Current sits between your code and the channels. One endpoint scores, routes, and reroutes on every request, so your inference takes the path of least resistance and your stack never changes.

One channel

Integrate once. Reach the whole market.

Point your existing OpenAI SDK at Current and every provider flows behind one key and one bill. No new client, no per-provider plumbing.

·
Providers behind it
·
Models served
·
Cheapest-path savings
channels in one out

scored every request · routed to the winner

Smart routing

Every competing offer gets scored on five axes (price, latency, uptime, liquidity, and health). The highest score carries the request, and we hand you the full breakdown.

Automatic failover

If an offer errors or rate-limits, Current reroutes to the next best one before a single byte streams. The current always finds a way through.

Marketplace pricing

Providers compete to serve each model, so you pay the winning offer's price. That's the floor, no markup. Current keeps a flat 5% out of the provider's settlement, never added to your bill.

The routing engine

Every request is scored,
then carried.

No black box. Every competing offer earns one weighted score across cost, latency, uptime, liquidity, and health, and the current shows you exactly why it chose the one it did.

score()highest wins
# for each channel serving the model
score = w_cost·cost
+ w_latency·latency
+ w_uptime·uptime
+ w_liquidity·liquidity + w_health·health
# each term min-max normalized 0..1
# the highest score carries the request
Default weightstunable per key · per request
Cost40.0%
Latency20.0%
Uptime20.0%
Liquidity10.0%
Health10.0%
Live channel ranking

Scored against the network right now

Cost Latency Uptime · liquidity · health

The channels

Every channel the current can take

Each provider's offers are channels. For a request we score every active offer that can serve the model on price, latency, uptime, liquidity, and health. The request goes down the highest scorer, and the rest stay queued so the current reroutes the moment one fails.

0 / 0 channels openLatency, uptime, and a 0 to 100 health score update live from real dispatch outcomes and feed straight into the routing engine.

OpenAI-compatible API

Change one line. Keep your code.

Point your existing OpenAI SDK at the Current. One base URL, one key, and every model becomes a channel the routing engine can reach.

quickstart.pylive
from openai import OpenAI

client = OpenAI(
    base_url="https://api.currentinference.com/v1",
    api_key="cur_live_••••",
)

# Same SDK. The Current finds the best channel.
resp = client.chat.completions.create(
    model="llama-3.3-70b",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(resp.choices[0].message.content)
Exposed surface
  • /v1/chat/completions
  • /v1/completions
  • /v1/embeddings
  • /v1/models

Pricing

You pay the marketplace floor price. No markup.

No subscriptions, no marked-up tokens. Providers compete to serve each request and you pay the winning offer's price. Current takes a flat 5% marketplace fee out of the provider's settlement. It's never added to your bill.

You pay
$0.00/Mtok

the winning offer’s price, per 1M tokens

Offer price (you pay)
$0.60
the competitive floor · routed now
=
Seller settlement
$0.57
95% to the provider
+
Marketplace fee
$0.030
flat 5% · our only revenue

Pay by card (Stripe)·USDC top-ups coming soon·Free cur_test_ sandbox keys

Roadmap

From one channel to the default routing layer

One current, flowing downstream. Each phase carries the network further toward the default routing layer for open-model inference.

  1. phase_01shipping

    One channel

    One OpenAI-compatible endpoint over every channel. It's backed by a live model catalog and routing you can actually see through.

    • OpenAI-compatible API
    • Model catalog
    • 5-axis routing
    • Console dashboard
  2. phase_02shipping

    Routing that learns

    The current adapts. Routing picks the highest score across price, latency, uptime, liquidity, and health. Failover reroutes around real offer health, and you get analytics on every request.

    • 5-axis scoring
    • Pre-stream failover
    • Health scored 0 to 100
    • Routing analytics
  3. phase_03shipping

    The marketplace is open

    Add channels of your own. Sellers self-onboard, list competing offers from their own endpoints, and earn settlement on every request they carry. They keep 95%.

    • Seller onboarding
    • Competing offers
    • Two-sided settlement
    • Seller payouts
  4. phase_04

    The network layer

    The default routing layer for open-model inference: on-chain seller payout rails, multi-region scale, and decentralized reputation for every offer.

    • On-chain payouts
    • Multi-region scale
    • Decentralized reputation

Inference takes the path of least resistance.

One channel. Lowest cost. A current that always finds its way through, across every provider, on every request.