OpenAI makes everyday AI agents much cheaper to run

OpenAI is turning GPT-5.6 from a model launch into a pricing and operations story. The company says Luna, its fastest and cheapest GPT-5.6 tier, now costs far less, while Terra, the balanced everyday tier, also gets a price cut. Those reductions also affect how usage is counted in ChatGPT Work and Codex, so teams can run more routine tasks without burning through the same budget.

What changed #

Luna is positioned for high-volume work: classification, document handling, routine coding steps and agent loops where cost matters as much as raw intelligence. Terra sits in the middle for everyday knowledge work and scoped tasks. Sol remains the top tier, but API customers now get Fast mode for latency-sensitive jobs. It replaces the older Priority Processing option and keeps the same model intelligence at a higher speed tier.

OpenAI also says the gains come from model, inference and workflow efficiency improvements, including production optimizations helped by GPT-5.6 Sol itself. That matters because lower serving cost can flow directly into cheaper product usage instead of staying as an internal infrastructure win.

Why it matters #

For buyers and builders, this is the practical side of frontier progress: useful AI gets cheaper when strong smaller tiers can handle more of the work. A team can route ambiguous planning or review to Sol, then hand clearer implementation and checking steps to Luna or Terra. If the quality holds, the same budget supports more agents, longer workflows and more automation inside tools people already use.