OpenAI slashes GPT-5.6 Luna prices by 80%, putting a dollar task at 6 cents

OpenAI is taking a cleaver to its pricing. The company announced that starting July 30, its most affordable model, GPT-5.6 Luna, will see an 80 percent price cut, dropping to $0.20 per million input tokens and $1.20 per million output tokens. The mid-tier Terra model isn’t untouched either, getting a 20 percent haircut to $2 and $12 respectively. The top-end Sol model holds steady. To put that Luna drop in perspective, OpenAI claims a task that cost a buck on leading models just a year ago now runs about 6 cents on Luna and executes nearly nine times faster. It’s an aggressive move that feels less like a routine update and more like a shot across the bow in an escalating price war.

This isn’t simply a PR stunt funded by a fresh round of venture capital. The company points to an unexpected source for the savings: its own AI. OpenAI says GPT-5.6 Sol essentially optimized its own GPU software, slashing deployment costs by 20 percent. It also reportedly wrung out more than a 15 percent improvement in token generation speed through speculative decoding. If an AI is now materially reducing the cost of running itself, the economic flywheel here is genuinely new and worth watching.

But let’s not pretend this is happening in a vacuum. The market is suddenly flooded with cheap, capable alternatives, particularly from Chinese firms applying relentless pricing pressure. Even OpenAI’s biggest backer, Microsoft, is now openly hawking its own MAI models as a more cost-effective alternative to the company it bet billions on. That’s an awkward dinner conversation waiting to happen. The cuts are clearly a defensive move to keep developers from decamping for cheaper options.

There’s a real risk lurking beneath the price drops. Frontier labs are tied to eye-watering infrastructure investments, and if this race to the bottom suppresses revenue growth too quickly, it could shake the balance sheets underpinning the entire sector. Cheap access is fantastic for builders, but nobody should feel comfortable if the labs building the most advanced systems can’t afford to keep the lights on.