2026-10-27: API 1.39

Downloads, responses and PDFs up to 64 MB, model tokens priced per model, and PDF tables that keep their headings.

#Added

• api: Downloads, http responses and PDFs may be up to 64 MB, from 20 MB: download.max_bytes, max_response_bytes, an agent browser's downloads and the PDF gate. A larger file is still refused, a PDF with pdf_too_large.
• dsl: read_pdf takes min_table_rows (1 to 20, default 3), the fewest lines with two or more cells a table needs.
• pricing: Kaabist's model's tokens are priced per model, input and output apart, at that model's own rates (plans and pricing). Thinking tokens count as output.
• pricing: A heal of a parked job, a canary or a drift signal that asks Kaabist's model is charged to the organisation by its tokens, as an llm line on the job it looked at. Test-mode jobs and your own models are not charged.

#Fixed

• dsl: read_pdf keeps up to three single-cell lines inside a table, such as section headings, as rows of it; drops a row of column numbers next to the header; and reads a header whose titles run into each other.