If you've used APIs from more than one LLM provider, you've probably hit this: switch a model provider, and you need to change endpoints, auth methods, and request formats. OpenAI uses a messages array, Claude uses messages plus system prompt, Gemini is yet another format.
new-api unifies all of that.
33,573 stars on GitHub, 7,406 forks, up 1,978 stars this week. Written in Go with 5,741 commits — a very actively maintained project.
Core features
Format conversion. new-api converts various LLM APIs into OpenAI-compatible, Claude-compatible, or Gemini-compatible formats. Your code writes once; switching models means changing config in new-api, not rewriting code.
Model aggregation. Connect multiple model providers (OpenAI, Anthropic, Google, domestic providers) through a single unified gateway.
Quota and billing management. Set different quotas, rate limits, and model access permissions per user/team. For individuals, manage balances across multiple purchased models. For enterprises, handle internal API distribution and cost control.
Load balancing and failover. Configure multiple providers as backups for the same model, with automatic failover.
Relationship to one-api
People in this space may know the one-api project. new-api can be viewed as an evolution of one-api with significant improvements: better UI, more complete billing system, broader model support, and more flexible plugin mechanisms.
5,741 commits and 658 tags show extremely high iteration frequency.
Use cases
Individual developers saving money. Aggregate OpenAI credits, Google free quotas, domestic provider new-user offers into one entry point. Use the cheapest option per scenario.
Team API management. Instead of 10 people each buying APIs, centrally procure and distribute through new-api. Track who used what and which model cost how much.
Application development. Build new-api as a middleware layer — adding or switching models never requires changing business code.
My take
new-api's core value is reducing switching costs.
In 2026's fiercely competitive model landscape, today's best model might be surpassed tomorrow. new-api adds an abstraction layer between your business logic and the model layer — switching models means changing config, not code.
33,573 stars prove this is a widely recognized need. If you use multiple model APIs, new-api is nearly essential infrastructure.
Sources:
- GitHub - QuantumNous/new-api
- GitHub Trending Go Weekly
- Project documentation and releases