OpenAI Launches GPT-6 Sol and Luna in ChatGPT Work and Codex to Cut Developer Inferencing Costs

OpenAI expanded its GPT-6 ecosystem by introducing GPT-6 Sol and GPT-6 Luna across ChatGPT Work and Codex environments, following the earlier debut of its flagship GPT-6 Astra model. These additions are designed to give enterprise teams and developers selectable reasoning tiers separated from the default chat interface, with Sol aimed at complex coding and agentic workflows, while Luna serves focused, repeatable, high-volume tasks at significantly reduced token prices compared to GPT-5.6 predecessors. For DevOps leads and engineering managers, this model segmentation is vital. As teams embed AI deeper into continuous integration pipelines, IDE environments, and task automation, default large models like Astra become economically impractical for routine code generation, refactoring, and automated status handling. Introducing Luna and Sol directly into Codex with adjustable reasoning effort controls allows organizations to fine-tune unit economics, preventing developer seat costs and automated task budgets from ballooning while maintaining high performance on specialized coding assignments. This update fits a broader, industry-wide shift from monolithic model selection toward tiered, workload-specific model routing. Leading AI labs have faced capacity crunches and high inference overhead when users default to top-tier frontier models for mundane tasks. By segmenting the GPT-6 line into Astra (heavyweight reasoning/science), Sol (mid-tier agentic work), and Luna (lightweight, rapid iteration), OpenAI mirrors architectural strategies used in modern distributed systems, enabling downstream tooling to dynamically target the appropriate balance of speed, cost, and reasoning depth. In practice, engineering organizations using ChatGPT Work and Codex should review their workspace defaults and automated tasks. Workspace administrators on Enterprise and Business tiers need to actively enable the models and set governance policies around reasoning effort levels. Platform teams should systematically test GPT-6 Sol for complex multi-file pull request reviews and migrate routine linting, unit-test authoring, or script maintenance to GPT-6 Luna to capture immediate token savings before older GPT-5.5 endpoints face scheduled retirement.