LegalOn says selective Codex use cut estimated costs by 65%

LegalOn says it cut estimated daily Codex costs by roughly 65% without slowing development, according to a vendor published OpenAI customer case study published October 8, 2026. For software teams watching AI bills, the useful detail is how: LegalOn changed model selection, restricted Fast mode and set spending limits together.

Developers had initially received unlimited access to GPT-5.5 in Fast mode. Under the revised approach described by OpenAI, GPT-6 Luna handles clearly specified implementation work, GPT-6.1 Sol takes standard design and analysis, and GPT-6 Astra tackles complex architecture decisions. Teams also ran tasks in parallel to preserve development speed after Fast mode became an exception.

OpenAI’s pricing announcement distinguishes seats billed by token consumption from standard Business seats with included Codex usage limits. The case study doesn't specify LegalOn’s seat arrangement.

Savings are not customer value

The case study gives no dollar baseline, measurement period or numerical benchmark for development speed. Its savings figure is an estimate against GPT-5.5 usage, and several changes contributed, leaving each change’s contribution unclear. Although its title says costs were halved, the detailed result is an estimated 65% reduction, with no explanation for the difference. That makes the result a useful example to investigate, rather than a savings forecast for another team.

LegalOn is still building a system to connect AI spending with the customer value of each feature release. That’s the next result to watch: whether faster releases deliver measurable gains for customers.