Their announcement here: https://openai.com/index/advancing-the-price-performance-frontier-with-gpt-5-6/
Their goal was never about the high price; it was to keep us from using the new models all along.
BYOK is available for that price in cursor last I heard, so I don’t think you’re far off here from what I know about the product.
The token model irks me too. We all have to give an estimate of cost and outcome before we start work. The only exception is lawyers.
Cursor’s “request” model made this palatable, but the rug’s been pulled where we’re down to Opus 4.6 and Codex 5.3 (and even then we don’t know how many requests will be charged per request) to run plans and need a separate Claude Code account to generate those plans.