Now in public beta · drop-in OpenAI replacement

One API. The cheapest model every time.

Promptoll routes every OpenAI-compatible request to the most cost-effective model without sacrificing quality. Multi-model arbitrage, token-accurate billing, global delivery.

zero changes — just a new base_url
from openai import OpenAI

client = OpenAI(
    base_url="https://promptoll.com/v1",
    api_key="zai_•••"
)

client.chat.completions.create(
    model="auto",     # Promptoll picks the cheapest
    messages=[{"role": "user", "content": "Hello"}]
)

Model routes

30+

across OpenAI, Anthropic, and domestic providers

Compute integration

Available

compute launch, lifecycle, and status dashboards

API compatibility

OpenAI

drop-in SDK calls are supported

Backed by every major model
OpenAIAnthropic ClaudeDeepSeekGoogle GeminiQwenKimi

30+ models, one API

Every one the provider's official full version

ModelInput / 1MOutput / 1M

Everything you need to ship AI at the right price

The OpenAI API on the outside. A cost optimization engine on the inside.

Smart routing

Send model="auto" and we pick the cheapest model that meets your quality bar. Send quality="high" for the flagship.

Model arbitrage

Ask for GPT-4o, get the same answer from a model costing 90% less — transparent, opt-in, per-tenant.

📊

Token-accurate billing

Every prompt and completion token logged at microcent precision. See cost, price and margin per request.

🔌

OpenAI + Anthropic compatible

Drop-in replacement for both the OpenAI SDK and the Anthropic Messages API (tool use, vision, extended thinking included) — change the base_url and you're done.

🔑

Multi-tenant keys

Issue scoped zai_ keys, set per-tenant markup, revoke instantly. Ideal for reselling AI inside your product.

🌐

Global delivery

Backed by a global edge network — predictable latency for users anywhere in the world.

Full IaaS capabilities

Compute infrastructure, not just a host toggle

VPC isolation, snapshots, custom images, independent data disks, elastic bandwidth — the same depth you'd expect from a dedicated cloud provider, running underneath your AI gateway.

See compute capabilities

Full visibility into every request

Every request shows requested vs. actual model, token usage, and what you were charged. Audit the full arbitrage effect — no hand-waving.

Model requested
Model used (after arbitrage)
Your charge
./request_logs.json
{
  "model_requested": "gpt-4o",
  "model_used":      "deepseek-v3.1",
  "prompt_tokens":    412,
  "completion_tokens": 189,
  "cost_microcents":   23940,
  "price_microcents":  86450,
  "latency_ms":        612,
  "stream":            true
}

From OpenAI to Promptoll in 60 seconds

Three steps. Zero lock-in. Point back to OpenAI any time.

01

Sign up

Create an account — no credit card. Get a zai_ API key in seconds.

02

Swap one line

Point your OpenAI SDK at promptoll.com/v1. No code changes.

03

Save instantly

Use model="auto" or opt into arbitrage. Watch cost drop in the dashboard.

04

Use in production

Call it from production traffic and monitor with traceable logs and host state.

Ship AI with receipts you can show investors.

Start free. Pay only for what you forward. Cancel any time.