Highlight Token Budget for POST /v1/search
extraction_mode: "highlights" now accepts an optional highlights object with total_max_tokens, a cap on the number of highlight tokens across the whole response. It takes values from 1024 to 8192. Omit it and the default budget of 8192 applies, so existing requests are unaffected.
A 10-result highlights response is roughly 6,000–7,500 tokens by default. With total_max_tokens: 1024 it drops to under 1,000, which suits agents that call search in a tight loop. The budget is shared by every result, so pair it with a lower count when you want fewer, longer results.