Skip to main content
Moss Inference bills tokens and retrievals from your credits. This page covers credits, the moss/fastlane-1 prices, vendor models, your own key and caching.

​
Credits

Inference bills from credits on every plan, with no included allowance.
  • Buy credits on the Billing page of the Moss portal. The minimum purchase is $5.
  • Set auto-reload on the same page.
  • Only organization owners can buy credits.
When your credits run out, requests return 429 usage_limit_exceeded until you add credits.

​
The Moss model

Tokens are the usage counts of each reply. Input tokens include the retrieved facts.

​
Vendor models on the Moss key

A vendor model called without your own key bills your credits per token. Models lists the input, cached input, cache write and output prices of each model. Retrievals bill at the same price as on moss/fastlane-1.

​
Your own key

With your own key in X-Moss-Provider-Key, the vendor bills the tokens to your vendor account. Moss bills only the retrievals. See Bring your own key.

​
Prompt caching

Prompt tokens a model already holds from an earlier call bill at the cached input price.
  • On moss/fastlane-1, these are in the prefix cache, such as a repeated system prompt or the earlier turns of a conversation.
  • The Chat Completions field usage.prompt_tokens_details.cached_tokens counts them, and Responses uses usage.input_tokens_details.cached_tokens. The rest of the prompt bills at the input price.
  • Where a model lists a cache-write price, prompt tokens written to the vendor’s cache bill at that price.
The reply cache is separate. It answers a repeated request without a model call, and the reply reports 0 prompt tokens (usage.prompt_tokens on Chat Completions, usage.input_tokens on Responses). A reply-cache hit is free.

​
Retrievals

A retrieval is one request that searches your indexes. It bills the same on every model, your own key included.
  • The price is $0.50 / 1k retrievals.
  • A call that names no index has no search and no retrieval charge.
  • A reply-cache hit is not a retrieval.
Ground answers in your indexes covers choosing indexes.

​
Usage

Inference is not billed as session minutes. The Usage page in the Moss portal shows inference tokens per model and retrievals.