Nov 11, 2024
NanoGPT recently added Yi Lightning and is the only service that provides easy access to it. Yi Lightning and its sister models Yi Large, Yi Medium and Yi Spark are created by the Chinese company 01ai. The Yi models are remarkably powerful models that are difficult to access, so we're excited to offer cheap access to them.
Yi Lightning is the flagship model of 01ai, a Chinese company founded in 2023 lead by Kai-Fu-Lee. The name might ring a bell from his book AI Superpowers: China, Silicon Valley and the New World Order.
In the book Kai-Fu Lee predicts that China will leap-frog the US in AI research and development. Proving his own point, when OpenAI and others restricted access for Chinese users 01ai stepped in and offered access to their remarkably powerful yet cheap models.
The independent leaderboards on LMArena are often used as a benchmark to rank model performance. Yi Lightning scores on par with Claude 3.5 Sonnet here, which is all the more remarkable given that Yi Lightning is a fraction of the cost.
There's just one issue with the Yi models: they're not available to everyone.
Getting access to the Yi models is a bit of a process. It requires a Chinese phone number, and even then access is not straightforward. We believe in making all the models available to everyone and are happy to be able to offer access to all the Yi models including Lightning, getting rid of a major barrier for those outside China that want to try the model.
01ai offers a range of models. Yi Lightning is the flagship model, but with Yi Large, Yi Medium and Yi Spark there is a range of options. Yi Spark is even cheaper than Yi Lightning, though the performance is slightly worse. Yi Medium has a token limit of 200k which allows for huge inputs and longer conversations, while Yi Large is their largest model.
Yi Lightning is a general purpose model that is well suited for a wide range of tasks. It's powerful enough for most tasks, yet much cheaper than alternatives like Claude 3.5 Sonnet or ChatGPT 4o. Users have reported replacing their API calls to Claude or ChatGPT with calls to Yi Lightning, saving immensely on costs.
For those interested in doing so our API offers an OpenAI-compatible endpoint to Yi Lightning. What this means is that you can use any OpenAI-compatible SDK or API client to connect to this model by replacing the endpoint with https://nano-gpt.com/api/v1/chat/completions, and that for most purposes replacing the endpoint in an existing application is as simple as changing the endpoint and API key.
We've made it as easy as possible to get started with the Yi models through the web interface, easy-to-use API endpoints and the ability to add funds using a broad range of options. We'd love to see what you build. If you are building on these models or have feedback for us regarding the Yi models we'd love to hear from you!
