Pollio

One API for any AI model, served by an open market of GPUs. With private inference, only you can read your prompts and outputs.

How to begin

Buy inference

Sign up
Use Google, GitHub, X, or a Solana or Ethereum wallet.
Sign in with
Add credits
Top up with any coin. Credits work with every model and every node, public or private.
Get your key
Create an API key and point any OpenAI client at Pollio. For private inference, run pollio proxy.
POLLIO_API_KEY
pl-live- followed by your secret key

Create an account

Sell inference

Install
Download the node for Linux or Mac. pollio-node doctor tells you which requests your machine can serve.
pollio-node doctor
Public:eligible
Private:needs attestation
Link your account
Run pollio-node login and approve the device from your account.
PLLO-7K3Q
Approve on pollio.ai/device
Serve and get paid
Set your prices, start the node, and get paid in USDC for every token you serve.
Payouts
- Sep 2641.20 USDC
- Sep 1937.85 USDC
- Sep 1229.40 USDC
Paid out108.45 USDC

Download the node

Copperplate engraving of Aequitas, the Roman goddess of fair dealing, seated in a radiate crown, holding a balance level in one hand while coins pour from the cornucopia at her side.

Keep your code, change the URL

Any OpenAI client works. Most models cost well below their makers' prices.

Fig. I, prompt

Agent setup

Set POLLIO_API_KEY in your shell, then paste this into Codex, Claude Code or OpenCode.

Set up Pollio as your model provider. Its API is at https://api.pollio.ai/v1 and the model is deepseek/deepseek-v4.1-flash. My key is in the POLLIO_API_KEY environment variable: never print it or write it into a file.+ 5 more lines
Codex: in ~/.codex/config.toml, add [model_providers.pollio] with base_url set to that URL, env_key = "POLLIO_API_KEY" and wire_api = "responses". At the top of the file, set model to the model and model_provider = "pollio".
Claude Code: in ~/.claude/settings.json, set env.ANTHROPIC_BASE_URL to that URL without /v1, set ANTHROPIC_MODEL, ANTHROPIC_DEFAULT_OPUS_MODEL, ANTHROPIC_DEFAULT_SONNET_MODEL and ANTHROPIC_DEFAULT_HAIKU_MODEL in env to the model, and set apiKeyHelper to "printenv POLLIO_API_KEY".
OpenCode: in opencode.json, add a "pollio" provider with npm "@ai-sdk/openai-compatible", options.baseURL set to that URL, options.apiKey set to "{env:POLLIO_API_KEY}" and the model under models, then set "model" to "pollio/deepseek/deepseek-v4.1-flash".
Any other agent: say so, and send me to https://pollio.ai/integrations/.

Back up the file first, keep my other settings, show me the change, and tell me what to restart.

Fig. I, terminal

Agent setup

export OPENAI_BASE_URL=https://api.pollio.ai/v1
export OPENAI_API_KEY=pl-live-...
export OPENAI_MODEL=deepseek/deepseek-v4.1-flash

Fig. II, claude_desktop_config.json

MCP server

{
  "mcpServers": {
    "pollio": {
      "command": "npx",
      "args": ["-y", "@pollio/mcp"],
      "env": {
        "POLLIO_API_KEY": "pl-live-..."
      }
    }
  }
}

Fig. II, terminal

MCP server

claude mcp add pollio --scope user \
  -e POLLIO_API_KEY=pl-live-... \
  -- npx -y @pollio/mcp

Fig. III, hello.py

API reference

# pip install openai
from openai import OpenAI

client = OpenAI(
    base_url="https://api.pollio.ai/v1",
    api_key="pl-live-...",
)

completion = client.chat.completions.create(
    model="deepseek/deepseek-v4.1-flash",
    messages=[{"role": "user", "content": "Hello"}],
)
print(completion.choices[0].message.content)

Fig. III, hello.mjs

API reference

// npm install openai
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.pollio.ai/v1",
  apiKey: "pl-live-...",
});

const completion = await client.chat.completions.create({
  model: "deepseek/deepseek-v4.1-flash",
  messages: [{ role: "user", content: "Hello" }],
});
console.log(completion.choices[0].message.content);

Fig. III, main.go

API reference

// go get github.com/openai/openai-go/v3
package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
	"github.com/openai/openai-go/v3/option"
)

func main() {
	client := openai.NewClient(
		option.WithBaseURL("https://api.pollio.ai/v1"),
		option.WithAPIKey("pl-live-..."),
	)
	params := openai.ChatCompletionNewParams{
		Messages: []openai.ChatCompletionMessageParamUnion{
			openai.UserMessage("Hello"),
		},
		Model: "deepseek/deepseek-v4.1-flash",
	}
	completion, err := client.Chat.Completions.New(context.TODO(), params)
	if err != nil {
		panic(err)
	}
	fmt.Println(completion.Choices[0].Message.Content)
}

Fig. III, request.sh

API reference

curl https://api.pollio.ai/v1/chat/completions \
  -H "Authorization: Bearer pl-live-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v4.1-flash",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Fig. IV, connection settings

Chat app setup

Base URL: https://api.pollio.ai/v1
API key: pl-live-...
Model: deepseek/deepseek-v4.1-flash

Fig. V, terminal

Provider guide

curl -fsSL https://pollio.ai/install.sh | sh
pollio-node login
pollio-node serve deepseek/deepseek-v4.1-flash \
  --price-in 0.003 \
  --price-out 0.012

The lowest current offer on the network, in US dollars, and how far below the maker's own API price it is, with the discount over the last 30 days.
Model	Best price	Discount
Claude Opus 5.5Anthropic · 1M context$0.75 / M input$3.75 / M output	$0.75 / M input$3.75 / M output	85%, between 79% and 99% over the last 30 days
GPT-6 AstraOpenAI · 1.1M context$0.18 / M input$1.08 / M output	$0.18 / M input$1.08 / M output	94%, between 94% and 99% over the last 30 days
GPT Image 2.5OpenAI$0.004 / image	$0.004 / image	90%, between 88% and 99% over the last 30 days
GLM 5.3Z.ai · 1M context$0.04 / M input$0.13 / M output	$0.04 / M input$0.13 / M output	96%, between 91% and 99% over the last 30 days
Kimi K3Moonshot AI · 1M context$0.04 / M input$0.18 / M output	$0.04 / M input$0.18 / M output	95%, between 65% and 98% over the last 30 days

The lowest current offer on the network, in US dollars, and how far below the maker's own API price it is, with the discount over the last 30 days.
Model	Best price	Discount
Claude Opus 5.5Anthropic · 1M context$0.75 / M input$3.75 / M output	$0.75 / M input$3.75 / M output	85%, between 79% and 99% over the last 30 days
GPT-6 AstraOpenAI · 1.1M context$0.18 / M input$1.08 / M output	$0.18 / M input$1.08 / M output	94%, between 94% and 99% over the last 30 days
GLM 5.3Z.ai · 1M context$0.04 / M input$0.13 / M output	$0.04 / M input$0.13 / M output	96%, between 91% and 99% over the last 30 days
Kimi K3Moonshot AI · 1M context$0.04 / M input$0.18 / M output	$0.04 / M input$0.18 / M output	95%, between 65% and 98% over the last 30 days
DeepSeek V4.1 FlashDeepSeek · 1M context$0.003 / M input$0.012 / M output	$0.003 / M input$0.012 / M output	97%, between 91% and 99% over the last 30 days

The lowest current offer on the network, in US dollars, and how far below the maker's own API price it is, with the discount over the last 30 days.
Model	Best price	Discount
GPT Image 2.5OpenAI$0.004 / image	$0.004 / image	90%, between 88% and 99% over the last 30 days
FLUX.2 [pro]Black Forest Labs$0.0036 / image	$0.0036 / image	88%, between 80% and 94% over the last 30 days
Seedream 5.0 ProByteDance$0.0049 / image	$0.0049 / image	86%, between 82% and 99% over the last 30 days

The lowest current offer on the network, in US dollars, and how far below the maker's own API price it is, with the discount over the last 30 days.
Model	Best price	Discount
Qwen Audio 3.0Qwen$0.002 / minute	$0.002 / minute	83%, between 79% and 93% over the last 30 days
Seed Audio 1.0ByteDance$0.003 / minute	$0.003 / minute	80%, between 66% and 80% over the last 30 days
Voxtral SmallMistral AI$0.001 / minute	$0.001 / minute	75%, between 72% and 91% over the last 30 days

Specimen prices. Names and logos belong to their owners. No endorsement implied.

See every model

In the tools you use

Keep your editor, agent or chat app. Only the provider changes.

All 19 integrations

Fig. I, config.toml

Illustrative setup

model = "deepseek/deepseek-v4.1-flash"
model_provider = "pollio"

[model_providers.pollio]
name = "Pollio"
base_url = "https://api.pollio.ai/v1"base_url = "http://127.0.0.1:8787/v1"
env_key = "POLLIO_API_KEY"
wire_api = "responses"

Provider: Pollio
Path: PublicPrivate

Codex

Add Pollio as a provider in one config block.

Fig. II, environment

Illustrative setup

ANTHROPIC_BASE_URL=https://api.pollio.aiANTHROPIC_BASE_URL=http://127.0.0.1:8787
ANTHROPIC_AUTH_TOKEN=pl-live-...
ANTHROPIC_MODEL=deepseek/deepseek-v4.1-flash

Provider: Pollio
Path: PublicPrivate

Claude Code

Same terminal, a different provider.

Fig. III, Cursor Settings

Illustrative setup

# Cursor Settings › Models
OpenAI API Key
  pl-live-...
Override OpenAI Base URL
  https://api.pollio.ai/v1
Custom Models
  deepseek/deepseek-v4.1-flash

Provider: Pollio
Path: PublicPublic only

Cursor

Override the base URL, keep your editor.

Private runs through pollio proxy on your machine.

Public and private

Copperplate engraving of the two-faced god Janus: two bearded heads joined at the back, one looking left and one looking right, bound with a tied fillet.

Choose a mode for each request. Public is the default.

Public

Your request goes to the open market, where providers compete to serve it.

Served by: Any provider on the network
Who can read it: You and the provider serving it
Price: The market rate, usually the lowest
Good for: Most work

Private

Prompts and outputs stay private. [To come: how private inference keeps prompts and outputs private.]

Served by: Providers that qualify for private requests
Who can read it: Only you
Price: Higher than public
Good for: Customer data and internal documents

For providers

Copperplate engraving of Vulcan seated on a rock at his anvil, raising a hammer over a bar of metal held in tongs.

If you run GPUs, you can sell inference on Pollio. Set a price for each model you serve and get paid in USDC every 7 days, less a 5% routing fee.

Read the provider guide

How a request travels

Buyer
Sends an OpenAI-format request, with an optional price cap and privacy mode.
Router
Picks a provider on price, latency and uptime. Private requests go only to providers that qualify.
Provider
Runs the model and streams the tokens back. If it stalls, the router moves the request.
Settlement
Counts the tokens, charges the buyer and pays the provider.

The router adds about 18 ms to a request.

Questions

Which models can I call?

Any model a provider serves on the network, 412 at the moment. If a model you need is missing, anyone can start serving it.

How is the price set?

Providers set their own price per million tokens. The router sends each request to the cheapest provider that meets your latency and uptime needs, and you can cap the price or pin a provider.

What if a provider fails mid-request?

The router moves the request to another provider and carries on. You only pay for the tokens you receive.

What do I need to become a provider?

A machine with a recent GPU and a steady connection. The node tells you which models fit in your card's memory, and you choose which ones to serve.

Sign up

Add credits

Get your key

Install

Link your account

Serve and get paid

Keep your code, change the URL

Discounts by model

Codex

Claude Code

Cursor

Public

Private

Buyer

Router

Provider

Settlement

Which models can I call?

How is the price set?

What if a provider fails mid-request?

What do I need to become a provider?