OpenAI Launches Cheaper GPT Models for Everyday AI Work

OpenAI has expanded its GPT-6 generation with Sol and Luna, two models positioned below the flagship Astra but aimed at making advanced AI cheaper to run in production. TechCrunch reports that the new models cut API prices compared with their GPT-5.6 predecessors while targeting fewer factual and coding mistakes.

What changed

Sol is designed for more demanding work such as coding and complex tasks. Luna targets high-volume jobs with clearer goals, including summarizing documents, extracting information, and answering short questions. OpenAI says the models benefit from improvements in caching and inference, allowing the company to offer the GPT-6 versions at roughly half the cost of the previous 5.6 series.

The company also says GPT-6 Sol makes about half as many mistakes as its predecessor on an internal factuality evaluation. Those claims are not independent benchmark results, so builders should treat them as release assertions rather than production guarantees.

The models are available through the ChatGPT API and Codex for most paid accounts, with Luna also rolling out to the desktop app and free users. OpenAI expects the broader ChatGPT rollout to happen gradually.

Why builders should care

The practical significance is less about a single benchmark score and more about model routing. Production systems increasingly need several capability tiers: a stronger model for planning or difficult code, a faster model for routine transformations, and a low-cost option for large batches.

That approach fits the operational reality of agentic workflows. A workflow can reserve higher-cost reasoning for the few steps that need it, while sending classification, extraction, formatting, and follow-up actions to a cheaper model. Teams should still pin model versions, track tool-call failures, and run representative regression tests before changing a production route.

OpenAI’s earlier push for shared AI standards, covered in our policy analysis, is also relevant here: lower prices will broaden usage, but stronger controls around permissions, evaluation, and observability become more important as more workflows move from experiments into routine operations.

Key takeaway: Cheaper capable models make multi-tier routing more practical. The advantage will go to builders who pair lower inference costs with disciplined evaluation and rollback controls.