OpenAI slashed GPT-5.6 Sol's Max-tier reasoning budget by 87% four days after launch with no notice, then called it experimental parameter tuning when developers revolted. The incident reveals a structural risk in building on AI APIs that most startup contracts don't account for.