Model pricing gains a long-prompt tier

What

Model pricing can now include a separate price for long prompts, used once a request's prompt passes a set number of tokens.

• The baked-in model catalog reads a new long_prompt pricing field, keyed by above_prompt_tokens, next to input, output, cache and web_search rates. The catalog version went from 1697 to 1760.
• When input, cache-read and cache-creation tokens together exceed abovePromptTokens, the longPrompt rates are used instead of the base rates.
• Per-model usage records add a longPrompt part holding the threshold and the costs worked out at that tier.
• The pricing line shows "in/out per Mtok" (per million tokens) and, when a long-prompt tier exists, adds "for prompts over N".

Why

Cost figures for very long requests can now reflect the higher rate some models charge above a prompt size, and you can see where that threshold sits.