LegalOn Technologies, a startup using AI for legal work, found that unlimited access to high-performance models threatened its budget. According to an OpenAI case study, the company initially gave developers unrestricted access to GPT-5.5 in Fast mode, but as adoption grew it needed a way to control spending without losing the productivity gains.
The solution was a set of guidelines for choosing among three models: GPT-6 Luna for straightforward implementation, GPT-6.1 Sol for standard design and analysis, and GPT-6 Astra for complex architecture and orchestration. The company also restricted Fast mode by default, allowed individual requests when needed, and relied on practices like running tasks in parallel to maintain development speed.
Budget caps were set at department, group, and individual levels. Mature business areas were asked to improve cost efficiency by up to about 20%, while newer businesses received generous budgets to encourage heavy AI use. Together, these measures cut estimated daily costs by roughly 65% compared with the previous approach, while keeping development speed intact.