# Amux AI vs LiteLLM: no servers to run, one account for every model

> LiteLLM is an MIT-licensed, self-hosted proxy: you deploy and operate it yourself and open accounts, configure keys and settle billing with each upstream provider separately, while data stays entirely in your own infrastructure. Amux AI is a hosted LLM API gateway: sign up and start, with one account and one balance for every model, no inference markup and low-cost channels far below official pricing and discounted official channels, plus native support for the Gemini protocol and Alibaba Cloud Model Studio, Volcengine Seedance, MiniMax and xAI image and video protocols, and enterprise-grade organization management.

Last verified 2026-10-19 · https://amux.ai/vs/litellm

- **0** Servers to run: LiteLLM needs PostgreSQL in production, and Redis for multiple instances
- **1 account** For every model: LiteLLM needs a separate account and key for each upstream provider
- **90%** Max saving vs. official price: From low-cost channels, based on live catalog prices; stable official channels are discounted too
- **4** Native video vendor protocols: Model Studio, Seedance, MiniMax and xAI — existing code only changes the base URL

## Amux AI vs LiteLLM, feature by feature

| Feature | Amux AI | LiteLLM |
|---|---|---|
| Deployment | Hosted — sign up and start, nothing to deploy or operate | Open source, self-hosted: PostgreSQL in production, Redis for multiple instances; the official docs describe self-hosted deployment |
| Upstream accounts and billing | One account and one balance for every model, with unified billing and invoices | You configure an API key for each upstream provider and settle billing with each one |
| Model pricing | No inference markup; low-cost channels go far below official pricing, and stable official channels are discounted too | The software is free; models are paid at each upstream’s official price |
| Text protocols | Native OpenAI Chat, OpenAI Responses, Anthropic Messages and Gemini (generateContent), with most models callable across protocols | OpenAI Chat, Responses, Anthropic Messages and Gemini generateContent |
| Image and video | Image: OpenAI Images and Alibaba Cloud Model Studio. Video: native Model Studio, Volcengine Seedance, MiniMax and xAI request formats plus a unified task API — existing client code only needs a new base URL and key | Unified-format image generation, image editing and video APIs |
| Embeddings and audio | Embeddings, speech and rerank endpoints (Coming soon) | Embeddings, speech, realtime, rerank, batches, fine-tuning and more |
| Multi-provider failover | 52 models are already served by multiple providers, with automatic failover and failed attempts not billed | Several routing strategies, fallbacks, retries and cooldowns — you connect and configure the upstreams yourself |
| Teams and budgets | Enterprise-grade organization management: organizations, workspaces and roles (including a finance role); spend limits per workspace, member and key | Budgets per key, team and user (database required); per-model budgets are an enterprise feature |
| Data control | Request and response bodies of successful text calls are not stored; customer content is never used for training | Self-hosted, so data stays in your infrastructure; no telemetry when self-hosted |
| Usage and export | Itemized charges for every call; monthly statements exported to Excel by key, model and workspace | Per-request cost in a response header; global spend reports are an enterprise feature |
| SSO and audit | SAML single sign-on and audit records (Coming soon) | SSO, audit logs and RBAC are enterprise features, priced by deployment size |

## Choose Amux AI if

- You would rather not deploy and maintain a proxy, database and cache, and want to start calling models right after signing up
- You do not want to open accounts, manage keys and top up with each upstream provider, and want unified billing and invoices
- You want leading models far below official pricing through low-cost channels, or at a discount through stable official channels, with automatic failover across providers
- You want to call image and video models in the native Alibaba Cloud, Volcengine, MiniMax or xAI request formats

## Choose LiteLLM if

- Data must stay in your own infrastructure, or you need an air-gapped or private deployment
- You already have cloud contracts or credits and want one call format on top of them
- You need a broader API surface such as embeddings, speech, batches and fine-tuning (embeddings and speech are on the Amux AI roadmap)
- You want to build on MIT-licensed code or deeply customize routing

## Switch from LiteLLM to Amux AI

Both work with the OpenAI and Anthropic SDKs. Switching only changes the endpoint, key and model names — your application code stays as it is, and there is no proxy left to maintain.

1. Sign up for Amux AI and create an API key in the console.
2. Replace the base URL of your self-hosted proxy with the Amux AI endpoint, and the key with your new API key.
3. Replace the custom model names from your proxy config with Amux AI manufacturer/model IDs, copied from the model page.

- OpenAI SDK base URL: `https://gateway.amux.ai/v1`
- Anthropic SDK / Claude Code base URL: `https://gateway.amux.ai`

```python
import os
from openai import OpenAI

client = OpenAI(base_url="https://gateway.amux.ai/v1", api_key=os.environ["AMUX_API_KEY"])
response = client.chat.completions.create(
    model="openai/gpt-6.1-sol",
    messages=[{"role": "user", "content": "Hello"}],
)
```

- Fallbacks and load balancing configured in your proxy are handled by automatic multi-provider failover in Amux AI; you can also set a routing strategy on the API key.
- Claude Code and the Anthropic SDK use the endpoint without /v1; the Gemini SDK can call generateContent directly.

## Frequently asked questions

### What is the main difference between Amux AI and LiteLLM?

LiteLLM is an open-source, self-hosted proxy: you deploy and operate it, and open accounts, configure keys and settle billing with each upstream, while data stays in your infrastructure. Amux AI is a hosted gateway: sign up and start, with one account for every model, no inference markup, low-cost channels far below official pricing and discounted official channels, native support for vendor image and video protocols, and enterprise-grade organization management.

### How can Amux AI price models below the official rate?

Amux AI connects each model to several providers in two kinds of channels: low-cost channels priced far below official pricing for cost-sensitive workloads, and official channels from model makers and major clouds that are stable and also discounted. Live per-provider prices are published on each model page; you can restrict an API key or workspace to official channels only, or set price-first routing.

### LiteLLM is free — will Amux AI cost more?

LiteLLM the software is free, but you pay each upstream at its official price and carry the cost of servers, databases and operations. Amux AI has no inference markup and no top-up fee, and low-cost channels go far below official pricing while official channels are discounted too, so the total cost is usually lower.

### What code changes are needed to switch from LiteLLM to Amux AI?

Usually three: point the base URL at the Amux AI endpoint, replace the key with an Amux AI API key, and replace the custom model names from your proxy with Amux AI manufacturer/model IDs.

### Does Amux AI store my prompts and responses?

Request and response bodies of successful text calls are not stored, and customer content is never used for training; prompts for image and video tasks are kept with the task for 30 days. Failed calls keep a limited sample for troubleshooting for up to 15 days, which you can turn off with a request header.

### Can Claude Code and similar tools switch from a LiteLLM proxy to Amux AI?

Yes. Amux AI natively implements the Anthropic Messages and OpenAI protocols, and setup guides cover Claude Code, Codex, Gemini CLI, Cursor, Cline, Cherry Studio and more — replace the endpoint and key as described.

### Can I put Amux AI behind a LiteLLM proxy?

Yes. Amux AI is OpenAI-compatible, so it can be configured as an upstream in LiteLLM, giving you one key and one balance for many models while keeping your proxy-level customizations.

_LiteLLM and other product names mentioned on this page are trademarks of their respective owners. Information about third-party products is compiled from their official public sources for evaluation purposes only and may change; refer to each vendor's website for current details. To report an inaccuracy, contact support@amux.ai._