According to an OpenAI case study published October 8, 2026, LegalOn Technologies reduced its estimated daily AI development costs by approximately 65% while maintaining development speed. The company, a legal-tech startup in the Asia-Pacific region, had integrated OpenAI's Codex into its development process and initially gave developers unlimited access to GPT-5.5 in Fast mode. As adoption grew, LegalOn reported rising costs and formed an internal group, the AI-powered Development CoE (AID CoE), to develop guidelines for choosing models by task. Under the new approach, teams start with a lighter model and move to more capable ones as complexity increases: GPT-6 Luna handles code implementation with clear requirements, GPT-6.1 Sol supports standard design and document preparation, and GPT-6 Astra handles advanced judgment such as architecture design. The company also restricted Fast mode by default and set monthly budget caps at the department, group, and individual levels. Established business units were asked to improve cost efficiency by roughly 20%, while new units received larger budgets. Senior Engineering Manager Yuta Tokitake said new businesses prioritized business speed over cost efficiency. The 65% figure is attributed to OpenAI and LegalOn and reflects estimated daily costs compared with GPT-5.5. The case study is a vendor-published account of its own customer.
- LegalOn reported an approximately 65% reduction in estimated daily AI development costs versus GPT-5.5
- Costs were cut by matching GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna to task complexity and development stage
- Mature business units were asked to improve cost efficiency by about 20%; new units received larger budgets
- Fast mode was restricted by default; monthly usage caps were set per department, group, and individual
- Source: OpenAI case study dated October 8, 2026
What it means for you
A company cut its AI spending roughly in half not by using less AI, but by stopping the habit of using the most powerful (and most expensive) model for every job. They used cheap models for simple tasks and reserved the expensive ones for hard problems. It's a vendor case study, so treat the exact 65% figure as the company's own claim, but the underlying principle is sound and applies to any business paying per-use AI bills.
Try this
Look at how you currently use AI tools and ask: am I defaulting to the most expensive model for everything? Try switching routine tasks (drafting, summaries, simple edits) to a cheaper, faster model and only reach for the top-tier model when the work genuinely needs deeper reasoning.
Who should care
Small businesses and teams whose AI usage has grown past the experiment stage and are now seeing meaningful monthly bills, especially anyone using coding assistants or per-token API models.
Skip this if
You're on a flat-rate plan, your AI usage is light, or you're not yet paying enough for AI to make cost management worth your time.
Sources: OpenAI — read the original