OpenAI Cuts Its Flagship Price and Undercuts Claude Opus 5

For most of this year the top-tier models from the big American labs held their prices steady while cheaper Chinese rivals piled in underneath them. That truce is over. On August 21 OpenAI cut the price of GPT-5.6 Sol, its most capable model, by more than 20 percent, and the new rate lands it below Anthropic's Claude Opus 5 on both ends of the meter.

Input tokens fall from $5 to $4 per million, and output tokens drop from $30 to $20 per million, according to OpenAI's pricing update. The company is calling the rate promotional and says it holds through at least November 21. It applies to API calls and to Codex and ChatGPT Work credits, so developers and enterprise customers both feel it.

The detail that matters is the comparison. Until this cut, Sol was the most expensive frontier model OpenAI sold. Now it costs less than Claude Opus 5, the model Anthropic positions as its general-purpose flagship. OpenAI was unusually direct about why, pointing to pressure from Anthropic and from a wave of low-cost Chinese models. When a company that has spent months refusing to discount suddenly names its competitors in a price announcement, that tells you how the market feels from the inside.

A discount that arrives from two directions

OpenAI is not the only one moving. Google recently launched Gemini 3.7 Flash at roughly half the cost of the model it replaces, pricing it at $0.75 per million input tokens and $3.75 per million output through the end of the year. DeepSeek went the other way with its V4-Pro release, raising peak-hour output pricing sharply, a reminder that not every lab is racing to the floor. The pressure on the American frontier, though, is coming from below, where open-weight coding models keep getting good enough to matter. Z.ai's GLM-5.3 and similar open releases now handle a large share of everyday coding and agent work at a fraction of frontier prices.

That is the squeeze. A lab can charge a premium for the smartest model in the world only for as long as the second-smartest model is meaningfully worse. As the gap narrows, the premium has to narrow with it. Cutting Sol below Opus 5 is OpenAI's bet that winning the developer who is comparing two spec sheets is worth more right now than protecting margin.

What it means for the people paying the bill

For anyone running these models at scale, the arithmetic is simple and welcome. A workload that leaned on Sol just got a fifth cheaper, and a team that had written off the top tier as too expensive might now run the numbers again. Output tokens are where most production bills concentrate, so the drop from $30 to $20 is the number that will move budgets.

There is a catch worth naming. The rate is promotional and tied to a date, which means today's comparison could reset in November. Anthropic has not responded with a cut of its own as of this writing, and it is worth watching whether it does. The company is preparing for a very large IPO, and public-market investors tend to reward pricing discipline as much as growth. A discount war is good for customers and hard on the story a company tells its shareholders.

Either way, the era of the frontier labs quietly holding the line on price looks finished. The competition has moved from who has the best model to who has the best model you can actually afford to run all day, and that is a healthier question for everyone buying.