Context Windows

Context window refers to the maximum number of tokens a model can process in a single request, including both the input (prompt) and the output (completion).

Token counting

Tokens are not equivalent to words. On average, one token equals roughly 0.75 words in English. A 1,000-word document is approximately 1,300 tokens. Code and structured data tend to tokenize at a higher rate than plain prose.

You can estimate token counts before making a request using the /v1/tokenize utility endpoint.

Managing long contexts

For documents that approach or exceed the context window, we recommend:

• Chunking the document and running multiple requests

• Using the crucible-2-preview model for extended context needs

• Summarizing earlier sections before passing them back into subsequent requests

Context and cost

Pricing is based on total tokens processed (input + output). Longer contexts cost more per request. For high-volume workloads with long documents, crucible-1-mini with chunking is often more cost-efficient than running full documents through crucible-1.