Unified LLM API Gateway 2026: LiteLLM, one-api and new-api Compared
Three open source gateways dominate: LiteLLM for the West, one-api and new-api for the Chinese ecosystem. Here's how to choose.
💡 What You Will Learn
Three open source gateways dominate: LiteLLM for the West, one-api and new-api for the Chinese ecosystem. Here's how to choose.
📜 Table of Contents
The Gateway Landscape
A unified LLM gateway gives you one API key and one endpoint for every model provider - plus routing, budgets, and logging. Three open source projects dominate 2026, and the right choice depends less on features (they overlap heavily) and more on your ecosystem: who maintains it, who else uses it, and where the docs are.
The Three
LiteLLM (56,266 stars, 2026-08-14) - the de facto international standard: a Python proxy exposing 100+ providers (OpenAI, Anthropic, Gemini, Azure, Ollama, vLLM, DeepSeek) behind one OpenAI-compatible endpoint. Massive community, enterprise adoption, and a config that is one YAML file. Best when your team works in English and needs the widest provider coverage.
one-api (36,372 stars) - the Chinese ecosystem's favorite: multi-provider proxy with a clean admin UI for distributing tokens to users, usage quotas, and per-channel management. Ideal for teams that want to give non-technical members API access without sharing master keys.
new-api (45,082 stars) - a one-api fork that outgrew it: more providers (including domestic Chinese models and image/video models), a better dashboard, billing/metering for reselling, and active development. The community pick for startups in the Chinese market.
The Decision Table
| Criterion | LiteLLM | one-api | new-api |
|---|---|---|---|
| Stars | 56,266 | 36,372 | 45,082 |
| Provider coverage | 100+ (global) | Many (CN-focused) | Most (CN + global) |
| Admin UI | Minimal | Good | Best |
| Billing/resale | No | Basic | Yes |
| Best for | Global prod | CN teams | CN startups/resale |
The 15-Minute Setup (any of them)
- Deploy via Docker (all three have images).
- Add your provider keys in the admin panel.
- Create user tokens/keys with quotas.
- Point your app at the gateway's OpenAI-compatible endpoint.
- Watch the logs: tokens, costs, errors per channel.
FAQ
Can I use two gateways at once? Yes - some teams run LiteLLM internally for engineering and new-api for user-facing key distribution.
Do they support local models? Yes - all three can proxy Ollama/vLLM endpoints alongside cloud providers.
Which has the best dashboard? new-api - it was built as a fork specifically to improve the UI and billing.
Is a gateway worth it for one developer? Marginal - the value appears at 3+ users or multi-provider setups; solo devs can start with direct APIs.
