# Infero

> The honest gateway for open models. Every open model behind one OpenAI- and Anthropic-compatible API key. The best price, tracked from the cheapest provider and only ever cut, with public price history and verified serving.

Infero is an OpenAI- and Anthropic-compatible inference gateway for open models (DeepSeek, Qwen, Kimi, GLM and the rest, for chat, image, speech and transcription), served through several upstream providers behind one API key. It is built by Elsa (https://heyelsa.ai).

Three commitments, each backed by a public endpoint:

- Tracker pricing: every model's price tracks the cheapest provider. Chat prices only move down automatically; provider cuts pass through and re-rate the current month. Image, transcription and speech prices change only when repriced. The full price history is public.
- Verified serving: every chat model and provider pair is canaried (tool-call conformance, code generation, unicode round-trip) and the scorecard is public. Requests that use tools never route to an endpoint failing the tool-call canary.
- Fair with money: prepaid credits that never expire, $0.50 free on first sign-in with Google, USDC and ELSA top-ups on Base credited 1:1 (ELSA earns 5% extra credit, up to $25 per account), per-key budgets and model allowlists.

## Docs

- [Full brief for language models](https://www.infero.xyz/llms-full.txt): this file plus the quickstart in full and the live lineup with prices, in one fetch.
- [OpenAPI description](https://www.infero.xyz/openapi.json): machine-readable description of the public and inference endpoints.
- [Quickstart](https://www.infero.xyz/docs): base URL, first request with the OpenAI and Anthropic SDKs, setup for Claude Code, Codex and OpenCode, and image, transcription and speech requests.
- [Pricing](https://www.infero.xyz/pricing): live prices per model: per million tokens for chat; per image, audio minute or million characters for media.
- [Status](https://www.infero.xyz/status): measured uptime and latency per provider, and the verified-serving scorecard.
- [Terms of service](https://www.infero.xyz/terms)

## API

Base URL: https://api.infero.xyz. Authenticate with `Authorization: Bearer sk-infero-...`. No key is needed for the public endpoints below.

- [Model list](https://api.infero.xyz/v1/models): OpenAI-format model list; `new: true` marks models added in the last 30 days.
- [Price board](https://api.infero.xyz/v1/pricing): the price per model, with every route's provider disclosed.
- [Price history](https://api.infero.xyz/v1/pricing/history): every price observation, oldest first; `?format=csv` for auditors.
- [Quality scorecard](https://api.infero.xyz/v1/quality): canary pass rates per model, provider and check over the last 7 days.
- [Status](https://api.infero.xyz/status): request outcomes and latency per provider and route, 24 hours and 7 days.

Endpoints for requests: `POST https://api.infero.xyz/v1/chat/completions` (OpenAI dialect, streaming supported) and `POST https://api.infero.xyz/v1/messages` (Anthropic dialect, tools and streaming supported), plus the OpenAI media endpoints `POST https://api.infero.xyz/v1/images/generations` (billed per image), `POST https://api.infero.xyz/v1/audio/transcriptions` (multipart, billed per second of audio) and `POST https://api.infero.xyz/v1/audio/speech` (billed per character). Chat models answer on both chat dialects; each image, transcription or speech model only on its own endpoint, which its `kind` in the model list names. The `x-infero-provider` response header names the upstream that served each request; send an `Idempotency-Key` header to make retries billing-safe.

## Optional

- [Sign in](https://console.infero.xyz/login): email one-time code or Google; signing in creates the account.
- [Elsa](https://heyelsa.ai): the company behind Infero.