A workaround for stale LiteLLM model pricing
LiteLLM can load a model cost map from GitHub to estimate each model’s price and . That shared file is maintained by the community, so new models can appear late, stay missing, or carry wrong prices until someone fixes them. One practical fix is to override prices model by model, or point `LITELLM_MODEL_COST_MAP_URL` to a separate cost map.
The example service at `cloudprice.net` uses the same format and refreshes prices from providers every day. It covers chat and embedding prices, plus image, audio, video, rerank, and OCR pricing. It also includes pricing for about 340 models that are not in LiteLLM’s default map, mostly newer releases such as Z.AI GLM-5.2, , and Vercel-related models.
The service is free, needs no key, and allows browser access, but has to prevent heavy use. A suggested product improvement is to let the Lite UI accept several cost map sources, using fallback sources when the first one lacks a model.
Key points
- LiteLLM’s default cost map can lag because it is community maintained.
- `LITELLM_MODEL_COST_MAP_URL` can point LiteLLM to a custom cost map.
- The example cost map refreshes provider prices every day and keeps the same format.
- It covers chat, , image, audio, video, rerank, and OCR pricing.
- It adds pricing for about 340 models missing from LiteLLM’s default map.