OpenAI's GPT-5.6 lands today, three versions deep, and it's already undercutting itself on price

OpenAI just did something it rarely does: undercut its own flagship model on price before most people even got to use it.

GPT-5.6 arrived in three tiers, Sol, Terra and Luna, rolling out from limited preview toward general availability over the coming weeks. Sol is the flagship, priced at $5 per million input tokens and $30 per million output tokens. Terra sits in the middle at $2.50 and $15, offering performance competitive with GPT-5.5 for roughly half the cost. Luna is the budget option at $1 and $6, built for high volume, lower stakes work.

That Terra pricing is the real story. A model matching last generation's flagship performance at half the price is not a minor iteration, it is OpenAI actively pressuring its own older, more expensive models out of the market before anyone else gets the chance to.

The timing is not an accident either. Sol is launching on Cerebras hardware this month at up to 750 tokens per second, a genuinely fast serving speed, right as Anthropic keeps eating into OpenAI's market share with Claude's own recent releases. When your biggest competitor is winning developers over on speed and value, cutting your own prices and chasing raw speed simultaneously is the obvious move, and OpenAI made it fast.

Access is still the sticking point. GPT-5.6 launched to a limited group of trusted partners through the API and Codex only, not the general public, so most people reading about Sol, Terra and Luna cannot actually touch them yet. General availability is promised for the coming weeks, which in AI launch speak can mean anywhere from days to months.

The AI price war has quietly become the actual competition to watch, more than any single benchmark chart. Whoever gets genuinely fast, genuinely cheap models into developers' hands first wins the next six months of mindshare. Right now OpenAI is racing hard to make sure that is still them.