AIQTry now
InputOutputGateway

LLM cost studio

Measure, route, and review what every model call costs.

AIQ estimates input, output, processing, and total spend, then lets you probe a live agent, publish a shareable gateway, and attach the security products you already license.

  • ShapePrompt, system, data, files
  • ReplyChat, stream, or thread
  • ScaleUsers, visits, calls a day

Drag a call across the studio

Input

Prompt

The words you send

Input

System

Instructions that ride along

Input

Data

Rows, context, retrieval

Input

Files

Attachments on the call

Output

Reply

Chat, stream, or thread

Overhead

Processing

Counted in the total

Total

Per call

Input + output + processing

Scale

Calls / day

Users, visits, volume

Scale

30 days

The month, from the same rate

Route

Gateway

Latency, failover, or weight

Route

Audit

Requests, errors, p95, cost

Cover

Security

Watch or block a surface

The studio

Dashboard

One live view of daily spend, the last probe, gateway traffic, analytics findings, and which security licenses are connected.

Measure

Cost a prompt, system prompt, data points, attachments, and multi-turn threads. Scale the same call by users, visits, and calls per day.

Endpoints

Save an agent URL — OpenAI Chat, OpenAI Responses, Anthropic, or custom JSON — probe it, and cost the usage the response returns.

Gateway

Publish saved endpoints as a shareable /api/g/{slug} proxy. Route by lowest latency, priority failover, or weight, with consumer keys, rate limits, and an audit log.

Analytics

Compare catalog models on the current workload and surface waste. Switch between the estimate and the latest live probe to see suggested savings.

Security

Attach a license you already have: Azure AI Content Safety, AWS Bedrock Guardrails, Google Cloud Model Armor, or Aikido. AIQ does not sell those products.

How a session runs

01

Shape the workload

Pick a catalog model and a conversation shape — chat, stream, or thread — then read input, output, processing, and total cost, plus a daily and 30-day projection.

02

Probe a live agent

Add an upstream URL on Endpoints. A successful run can be linked back into Measure and Analytics so the numbers follow real usage instead of the estimate alone.

03

Publish a gateway

Turn one or more saved endpoints into origins. Share a consumer key and call POST /api/g/{slug} with Authorization: Bearer <key>. Pause traffic without deleting the gateway.

04

Review and cover

Read findings on Analytics, then choose which surfaces — Measure, Endpoints, Gateway — an existing security license should watch or block.

What you need to know

Models
OpenAI, Claude, Gemini, and SpaceXAI. The Pricing Grid is the USD catalog per 1M tokens. Edit any cell in USD; the header currency converts what you read.
Currency
USD, EUR, GBP, or INR. Catalog rates stay in USD.
Keys
None are required to estimate cost. Bring your own only when probing an endpoint, publishing a gateway, or connecting a security provider.
Where data lives
The Measure session, saved endpoints, pricing edits, and theme stay in this browser. Published gateways and security connections stay on the server. Consumer keys are hashed; the plaintext key is shown once when issued.
Request safety
Probe and gateway requests refuse private, loopback, and metadata URLs.
Measure, route, and review what every model call costs | AIQ