Amux AI

One API for every major model.
Compare providers, pay the list price.

Routing overhead measured in milliseconds, automatic failover across providers, built for production traffic. Negotiated provider rates go straight into the unit price. Pay-as-you-go today, subscriptions coming soon.

Base URL
Get startedBrowse models

One-line integration

Change the base_url.
Call every model.

Text models are natively compatible with the three general-purpose protocols — OpenAI, Anthropic and Gemini — while image and video models also accept each vendor’s native API. No changes to your SDKs, agents or tooling.

  1. 1Keep your existing SDK and code
  2. 2Point base_url at Amux AI
  3. 380+ models, ready to call
main.py
import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.openai.com/v1",
    api_key=os.environ["AMUX_API_KEY"],
)

reply = client.chat.completions.create(
    model="anthropic/claude-opus-5-5",
    messages=[{"role": "user", "content": "Hello"}],
)

Connected · 80+ models available

Models, providers and agent clients that work out of the box

Callable models
80+
Manufacturers
13
Providers
14
Native API protocols
10+

Compare by provider

One model.
Every quote, best price first.

Live quotes for Claude Opus 5.5 from 5 providers. Prices for the same model can differ several-fold; Amux AI lists them side by side, and you can set price-first routing on an API key. Zero markup on inference — you pay the provider’s list price.

Anthropic list price
$20
  • Amazon KiroLowest
    Value tier
    -95%$1
  • Amux Special
    Value tier
    -85%$3
  • Amux AI
    Quality tier
    -70%$6
  • Anthropic
    Quality tier
    -30%$14
  • Amazon Bedrock
    Quality tier
    $20

USD per 1M tokens. Includes active promotions; synced with the model catalog every 5 minutes.

See all prices for Claude Opus 5.5

Transparent billing

Every charge,
fully traceable.

Pay-as-you-go, with subscriptions coming soon. Negotiated provider discounts go straight into the unit price; every charge is itemized, billed monthly and exportable.

Inference markup
0%
Billed at the provider’s list price
Top-up fee
0%
Every top-up is credited in full
Top discount
-95%
Deepest promotion currently live in the catalog
Sample billMonthly statement
Issued
  • Claude Opus 5.5$186.42
  • GPT-6.1 Sol$94.18
  • Seedance 2.5$48.60
  • Nano Banana 2.1$21.30
  • Account discount-$28.04
Total$322.46

Routing & failover

Every request,
on the optimal path.

  1. 01Smart routing

    Select a provider per request by price, latency, throughput or reliability. By default requests are spread across healthy providers; switch strategies on the API key, or, with per-request overrides enabled, append a suffix such as :@price or :@latency.

  2. 02Automatic failover

    When a provider errors or times out, the request fails over to the next one. No retry logic required in your code.

  3. 03Quality tiers

    Quality and value supply sit in separate candidate pools and are never mixed. Restrict a key or workspace to the quality tier and value supply is excluded from routing.

Quality tierValue tiertierScope: qualityYour appAmux AIAmazon Bedrock503Anthropic200 OKAmux Special

Every modality

Beyond chat.
Text, image and video. One key.

Every modality shares the same authentication, billing and request logs.

Long context, tool calling and thinking modes. Call the same model over the OpenAI, Anthropic or Gemini protocol.

Text-to-image, image-to-image and inpainting, priced per image or per token.

Text-to-video and image-to-video, submitted and polled as unified async tasks.

Claude Opus 5.5
OpenAIAnthropicGemini

Explain model routing in one sentence.

Model routing selects the best provider for each request automatically — lower cost, faster responses, and instant failover when a provider goes down.
POST/v1/chat/completions/v1/messages/v1beta/models/{model}:generateContent

Minimalist poster: mountains at sunset, soft gradients

Generate
Minimalist poster: mountains at sunset, soft gradients1024 × 768
4 imagesActual outputs from the same prompt
0%
Click for sound
0:00 / 0:15

Audio-guided rhythmUses the music as the timing reference — motion and cuts land on the beat and mood, making beat-synced edits effortless.

POST /v1/videos/generationsSucceeded

Built for teams

From side project
to organization.

Organizations, workspaces, member roles, key permissions and limits, and per-request logs — managed in one console.

Organizations & roles

Partition members, permissions and balances by organization and workspace. Invite members by email.

  • AOwner
  • LBilling
  • MAdmin
  • SMember

Per-key controls

Configure a model allowlist, supply tier and spend cap for every key.

productionamux-••••••••••••
Allowed models
claude-opus-5-5gpt-6.1-solgemini-4-argon+2
Quality tier only
Monthly cap$500.00

Request logs

Model, provider, latency, tokens and cost for every call, filterable by key and member.

  • claude-opus-5-51.2s$0.0184
  • gpt-6.1-sol0.8s$0.0061
  • gemini-4-argon2.4s$0.0123
  • deepseek-v4.1-flash0.4s$0.0004
  • kimi-k31.1s$0.0032

Limits & alerts

Daily, weekly and monthly spend caps for members and keys, with alerts before a threshold is reached.

This month$312.40 / $500
Today$18.90 / $25

Today's usage has reached 75%

Frequently asked questions

Amux AI is an LLM API gateway. A single API key gives you text, image and video models from OpenAI, Anthropic, Google, xAI, DeepSeek, Qwen and more. Each model is served by multiple providers, in low-cost channels far below official pricing and discounted official channels; Amux AI handles billing in one place, routes by your chosen strategy and fails over automatically when a provider has issues. The catalog grows continuously, and new models usually go live shortly after release.

One account and one balance cover every vendor — no separate sign-ups, top-ups or reconciliation. You can compare the same model across providers, with low-cost channels far below official pricing and official channels discounted too. If a provider is rate-limited or down, requests fail over to another provider and failed attempts are not billed. Teams can assign permissions by organization and workspace, and budgets by workspace, member and key.

Amux AI has no inference markup and no top-up fee, paid credit never expires, low-cost channels go far below official pricing, and official channels are discounted too. It natively supports the OpenAI, Anthropic and Gemini text protocols plus the Alibaba Cloud Model Studio, Volcengine Seedance, MiniMax and xAI image and video request formats, and it provides enterprise-grade organization management with budgets per workspace, member and key — no enterprise plan required. Side-by-side comparisons with OpenRouter, ZenMux, Vercel AI Gateway, Requesty and LiteLLM are on the Compare page.

No. Inference carries no markup and top-ups carry no fee — the credit you receive equals the amount you pay. Prices are also below official rates: low-cost channels go far below official pricing, and stable official channels are discounted too, as shown live on each model page. Our revenue comes from wholesale discounts negotiated with upstream providers.

Amux AI natively implements four text protocols — OpenAI Chat Completions, OpenAI Responses, Anthropic Messages and Gemini — and most models can be called across protocols (see each model page), so the OpenAI SDK can call Claude and Gemini too. Images are available through OpenAI Images and Alibaba Cloud Model Studio; video through the native request formats of Model Studio, Volcengine Seedance, MiniMax and xAI, plus a unified task API. With the official OpenAI, Anthropic or Gemini SDK you only change the base URL and API key. Setup guides cover Claude Code, Codex, Gemini CLI, Cursor, Cline, Cherry Studio and more.

By default, traffic is distributed with weights across healthy providers. You can set an API key to prioritize price, latency, throughput or success rate, or pin a specific provider in the request. On connection errors, timeouts, rate limits or server errors, the request moves to the next provider, and the platform absorbs the cost of failed attempts. Latency, throughput and uptime for every provider are published on each model page.

Request and response bodies of successful text calls are not stored; call records contain only metadata such as time, model, provider, usage, latency and status; prompts for image and video tasks are kept with the task for 30 days. Failed calls are sampled and kept for up to 15 days for troubleshooting only, and you can opt out with the header x-amux-omit-request: true. We never use your content to train models.

Cards are accepted, along with local payment methods in some regions — the checkout page shows what is available to you. Auto top-up is also supported. Every top-up comes with a receipt and invoice, with tax ID support. Unused paid credit can be refunded to the original payment method within 24 hours of the top-up, less the processing fee actually incurred for that payment.

Amux AI serves most countries and regions worldwide. It is not offered in mainland China, and it complies with applicable export controls and sanctions. See Section 15 of the Terms of Service.

Get started

Minutes from now,
your first request is live.

  1. 1Create an account
  2. 2Create an API key
  3. 3Swap the base_url and call
Base URL
Get startedView pricing