LegalOn Cuts Codex Costs by 65% While Protecting Speed
LegalOn Technologies successfully reduced its estimated daily Codex expenses by 65% without sacrificing development speed through strategic model selection and targeted budget management.
Radwan AbdulhadiEditor and technology writer
Navigating the Challenge of AI Cost Control
LegalOn Technologies offers Professional AI globally to support legal work and key business functions, combining domain expertise with artificial intelligence to help people focus on complex judgments and decisions. As the organization integrated Codex into its development process and expanded everyday adoption, a major challenge emerged regarding cost control. Unlimited use of high-performance models threatened to drive up spending, while blanket restrictions risked undermining the productivity gains that AI had enabled. Further details are available from OpenAI News in the original source material.
To reduce costs without slowing down development, the company selected among GPT-6 Astra, GPT-6 Luna, and GPT-6.1 Sol based on task complexity and development stage, while aligning budgets with each business unit's stage of growth. According to reports from OpenAI News, this comprehensive strategy successfully halved overall costs while maintaining the same development speed.
Establishing Guidelines and Model Optimization
Initially, developers at the company enjoyed unlimited access to the main model, GPT-5.5 in Fast mode, applying it broadly across design, implementation, and everyday work. However, continuing to use high-performance models without limits would inevitably exceed annual budgets. In response, the company's AI-powered Development CoE (AID CoE) began developing clear guidelines for model selection, testing and monitoring models so that managers could share findings with their teams.
This framework enabled individual engineers to independently choose the most suitable model for each task rather than defaulting to the highest-capability option for every assignment. Teams adopted a progressive approach, starting with lightweight models and advancing to more capable ones as task complexity increased.
Matching Models to Task Complexity
To support governance and refine selection criteria, LegalOn introduced administrator settings with monthly usage limits for both departments and individuals, while the AID CoE actively monitored and adjusted usage. Internal testing proved crucial in matching the right model to specific tasks across the organization.
Under the new structure, GPT-6 Luna acts as the lightest model, managing code implementation with clear requirements and everyday automations. Meanwhile, GPT-6.1 Sol supports standard software design, data analysis, routine numerical work, and document creation. For advanced tasks requiring complex analysis, design, and orchestration—including complex software design and coordinating agents—developers turn to GPT-6 Astra, the most capable model.
Tailored Budgets and Restricted Fast Mode
Alongside model selection, LegalOn restricted Fast mode by default, permitting individual requests only when strictly necessary. Although this shift initially sparked concerns regarding development performance, teams maintained their output by running tasks in parallel and transitioning smoothly.
The organization also implemented strict budget caps for AI spending at the department, group, and individual levels. Established LegalOn businesses were challenged to improve their cost efficiency by approximately 20%, whereas new businesses in the launch phase received generous budgets to encourage active AI use and business speed over short-term savings.
Measuring the True ROI of AI Investment
Yuta Tokitake, Senior Engineering Manager at LegalOn Technologies, noted that new businesses prioritized output to drive overall growth. Through the combination of model optimization, feature restrictions, and tailored budgets, the company successfully reduced estimated daily costs by approximately 65% compared to GPT-5.5.
Looking ahead, the company wants to understand whether faster development truly translates into customer value. Tokitake pointed out that while tracking the cost of individual tasks is straightforward, the total cost of a complete feature release often remains unclear. To address this, LegalOn is building a pipeline to measure ROI by linking the customer value of a release directly to the actual AI costs invested in it.
Sources
- OpenAI NewsLegalOn halves Codex costs while maintaining development speed
Continue chronologically
Related entity coverage
Related stories
Atlassian and OpenAI Expand Partnership to Power Enterprise AI
Atlassian and OpenAI expand their partnership, integrating frontier models to power AI experiences, Rovo, and developer tools like Codex.
OpenAI Releases Practical Guide for GPT-6 Integration
OpenAI details model selection, reasoning effort tuning, context optimization, and production best practices for the GPT-6 model suite.
NVIDIA GPUs Power OpenAI’s GPT-6 Astra Ultrafast Launch
GPT-6 Astra Ultrafast is now available, delivering up to 8x faster token generation on NVIDIA Blackwell GPUs for developers and enterprise users.
How Ultrafast Cuts Incident Investigation Time at OpenAI
Explore how OpenAI engineers use Codex Ultrafast to slash incident investigation times, improve security analysis, and streamline daily workflows.
