AI Tools for Business Get Cheaper as OpenAI Cuts API Prices in Half
OpenAI cut its API prices in half on September 22 by releasing two fresh models, Sol and Luna, both carrying the GPT-6 label, and the company says the new rates are permanent rather than a limited promotion.
Sol now costs $2 per million input tokens and $10 per million output tokens, down from $4 and $20 under the previous generation. Luna is even cheaper at $0.10 and $0.50 per million tokens. For a founder who has been rationing how much AI your team can actually use because of the bill, this is the kind of price cut that changes what you build next, not just what you pay for what you already run.
The Numbers Behind the Price Cut
The savings are not just about sticker price. Sol’s deception rate, a measure of how often the model produces confidently wrong answers, dropped to 1.3% from 10.4% in the prior version. Luna’s factual accuracy now roughly matches the older, pricier Sol model at a fraction of the cost per task.
| Model | Old input / output | New input / output |
|---|---|---|
| Sol | $4.00 / $20.00 | $2.00 / $10.00 |
| Luna | $0.20 / $1.20 | $0.10 / $0.50 |
OpenAI also improved prompt caching, cutting fresh-processing token needs by more than half across tools like GitHub Copilot. That detail matters more than it sounds, because cached tokens already carry a 90% discount, so the combined effect compounds for any product that reuses the same context repeatedly.
Why This Matters for Your Business, Not Just Your Bill
Founders who track AI spend control have spent this year watching model costs sprawl across product, engineering and marketing teams. A 50% price cut from the market leader forces every competing model provider to respond, and that competitive pressure benefits anyone buying AI capacity rather than selling it.
The bigger shift is strategic. OpenAI is explicitly encouraging businesses to route different jobs to different models: cheap, fast Luna for summarizing and extracting data, pricier Sol for coding and complex reasoning. That portfolio approach only makes financial sense once the cheap tier gets cheap enough to use constantly, which is exactly what happened this week.
What Founders Should Do Right Now
Audit which tasks your team currently sends to an expensive model out of habit rather than necessity. Simple summarization, tagging, or first-pass customer support drafts are strong candidates to move to a cheaper tier like Luna, freeing budget for the harder problems that still need a top-tier model.
This is also a good moment to revisit how your team measures return on AI spending in the first place. Sharper AI search optimization habits, paired with cheaper inference, mean smaller teams can now run experiments that used to require an enterprise budget. If you shelved an AI feature six months ago because the unit economics did not work, the math has likely changed.
What a Cheaper Default Model Means for Product Roadmaps
A lower price floor changes which features are worth building at all. Products that once required careful rationing of AI calls, such as running a model check on every customer message instead of a sample, suddenly pencil out financially. Founders who shelved an idea because the per-call cost made the unit economics ugly should revisit that math now, because the floor just moved.
There is a second, quieter effect. When inference gets cheap enough, teams stop treating AI calls as a scarce resource and start treating them as a default step in more workflows, similar to how cheap cloud storage changed how casually companies logged and kept data. That mindset shift tends to unlock features nobody planned for, simply because the cost of trying no longer feels risky.
Which Rivals Move Next
Watch how Anthropic and Google respond. Sol’s new pricing already undercuts Anthropic’s Opus 5.5, and Luna beats Google’s introductory rate for Gemini 3.8 Flash, so a further round of cuts within weeks would not be surprising. Founders locked into annual AI contracts should also watch whether their vendor passes savings through or pockets the margin.
Publishers are watching a related fight over how AI companies compensate the sites their models learn from. Recent moves around Google AI overviews payments suggest pricing pressure on the model side could eventually extend to content licensing too, which matters if your business depends on organic search traffic.
Frequently Asked Questions
Are the new GPT-6 prices temporary?
No. OpenAI confirmed the lower rates for Sol and Luna are permanent pricing, not a limited-time promotion tied to launch.
Which model should a small team use by default?
Luna fits high-volume, simple tasks like summarizing or tagging. Reserve Sol for coding and reasoning work where accuracy matters more than cost per call.
Does this affect tools built on the OpenAI API, like customer support bots?
Yes. Any product built on the API inherits the new pricing automatically, which lowers the cost of running high-volume features such as chat support or content drafting.
