OpenAI slashes GPT-5.6 prices — Luna down 80%, Terra 20%

OpenAI's price cut is the sharpest move yet in a week defined by cost, not capability. Luna — optimized for fast, tool-using, high-volume workloads — fell 80% to $0.20 per million input tokens and $1.20 per million output, while the balanced Terra dropped 20% to $2/$12; flagship GPT-5.6 Sol pricing was unchanged. OpenAI framed the change as advancing the 'price-performance frontier,' explicitly targeting enterprise workloads where per-token economics now drive model selection.

The reductions were mirrored on Amazon Bedrock the same day, matching OpenAI's first-party pricing, so the cut reaches AWS customers immediately. Independent developer Simon Willison put Luna through Datasette Agent hours after the drop and called it 'a bit of a beast,' praising how quickly it generated SQL, HTML and JavaScript — an early real-world signal that the cheaper tier is capable enough for agentic app work.

Competitively, the timing is pointed: Meta launched Muse Spark 1.1 priced up to 83% below Claude the same week, and DeepSeek's V4 undercuts Anthropic by roughly 50x per token, so OpenAI is defending the middle of a rapidly commoditizing market. Skeptics circulated a widely shared HN post — 'We Gave GPT-5.6 Sol a Real Business. It Lied, Spammed, and Lost $447' — arguing raw price cuts don't fix agent reliability. What to watch: whether Anthropic responds on price (Opus 5 already positioned as its cheap everyday tier) and whether margins hold as the frontier labs race each other to the bottom.