OpenAI takes GPT-5.6 public in three tiers (Sol, Terra, Luna), triggering a price war

GPT-5.6 is OpenAI's newest flagship family, now generally available after a staged rollout that began July 9. The three tiers are positioned by cost and latency: Sol for maximum reasoning ($5 input / $30 output per million tokens), Terra as the balanced default ($2.50/$15), and Luna as the fast, cheap option at $1/$6 — priced to undercut rivals and, per LlamaIndex's day-zero ParseBench run, delivering only 'minor degradations' versus Sol at roughly a sixth of the cost.

Sam Altman framed the launch around efficiency, posting that 'GPT-5.6 sol is half the price and ~twice as token efficient as fable in many cases for accomplishing the same task,' and that OpenAI is 'happy to deliver at one-quarter of the price.' Developers on ParseBench noted that GPT-class models still excel at tables and text but struggle with charts and complex layout — and that Luna's extra reasoning tokens don't translate to better visual understanding.

The competitive context is a full-blown price war: the LA Times reported OpenAI, Meta and Musk all slashing model costs the same week, while cheaper Chinese models like Z.ai's GLM-5.2 lure US firms. Anthropic responded by extending free Fable 5 access, and xAI's Grok 4.5 landed at $2/$6. Skeptics flag a 'paradox of choice' — the jumble of Sol/Terra/Luna names, tiers and reasoning levels confuses buyers — and Altman himself warned of reliability 'hiccups' as Sol demand outstrips inference capacity.