Open-source models
Open-weight models like GLM-5.3-Flash, DeepSeek-V4.1-Flash, GLM-5.3, and Kimi K3, through one OpenAI-compatible API. No GPUs to rent; pay per token, up to 30% below list prices.
Open-weight models
Each model here has official weights published by its vendor on Hugging Face. Prices are in US dollars per million tokens.
| Model | License | Input/M | Output/M | Context | Weights | Save |
|---|---|---|---|---|---|---|
GLM-5.3-Flashglm-5.3-flash | MIT | 1M | Hugging Face | 30% off | ||
DeepSeek-V4.1-Flashdeepseek-v4-1-flash | MIT | 1M | Hugging Face | 30% off | ||
GLM-5.3glm-5.3 | GLM-5.3 LicenseCustom license | 1M | Hugging Face | 30% off | ||
Kimi K3kimi-k3 | Kimi K3 LicenseCustom license | 1M | Hugging Face | 30% off |
Struck-through prices are list prices.
Why use them through an API
No GPUs to run
Frontier open models need a multi-GPU server to run. Here you call them like any other API and pay only for the tokens you use.
One key for every model
Every model shares the same base URL and API key. To switch models, change the model field; your OpenAI SDK code stays the same.
Prepaid, no subscription
Top up from $5 and pay per token. No monthly fee and no minimum spend.
About the licenses
Open weights means the vendor has published the model’s weights, so anyone can download, run and fine-tune it. Each model’s license is linked in the table above.
MIT-licensed models come with very few restrictions. Some vendors publish weights under their own license, based on MIT, that adds conditions for very large model-as-a-service providers. Read the license before you download or self-host the weights.