OpenAI and Anthropic slash AI prices in dueling releases 90 minutes apart

Anthropic and OpenAI, the two most closely watched companies in artificial intelligence, released rival model updates within roughly 90 minutes of each other Monday, both centered on the same pitch: similar capability to their existing top-tier models, at a sharply lower price.

Anthropic went first, announcing Claude Opus 5.5 with list prices cut 20 percent and typical workload costs down roughly 40 percent from its predecessor. OpenAI followed with GPT-6 Sol and GPT-6 Luna, two new models with API prices cut by as much as 50 percent compared with the versions they replace.

What happened

Opus 5.5 is the first entry in Anthropic's new "5.5" model family and, the company says, performs at roughly the level of its Claude Fable 5.1 model on most tasks while running about 30 percent faster than Opus 5. Anthropic priced it at $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5, with cached-token reads cut to $0.20 per million from $0.50. The model is available immediately through Amazon Web Services, Google Cloud, Microsoft Azure and Anthropic's own platform.

OpenAI's Sol, aimed at complex and agentic coding work, is priced at $2 per million input tokens and $10 per million output tokens, half of what the prior GPT-5.6 Sol model cost. Luna, built for high-volume, low-complexity tasks like summarizing or answering quick questions, costs just $0.10 per million input tokens and $0.50 per million output tokens, with cached inputs discounted a further 90 percent. Both models rolled out the same day across ChatGPT, OpenAI's Codex coding tool, the API, GitHub Copilot and Microsoft Foundry.

The numbers

Both companies backed the price cuts with performance claims rather than presenting them as pure discounts. Anthropic said Opus 5.5 scored 66.4 percent on the Terminal-Bench 4.0 benchmark and 81.8 percent on OSWorld 2.0's computer-use test, and cited customers including the legal-tech company Clio and the trading firm Optiver reporting it matched or exceeded prior-model quality in less time. OpenAI said Sol makes about half as many mistakes as GPT-5.6 Sol and reaches reliability close to its flagship Astra model at a fraction of the cost; on the same OSWorld 2.0 benchmark, OpenAI put Sol's score at 60.5 percent, essentially matching Claude Opus 5's 60.3 percent.

The two companies are "engaged in a price war" that is cutting into the future pricing power and profitability both firms will eventually need to recoup, said Ara Kharazian, lead economist at the corporate-card and spend-management firm Ramp.

How we got here

The dueling releases land two months after Anthropic's Opus 5 launch in July and follow a string of increasingly frequent model updates from both labs this year. Industry analysts have pointed to two converging pressures: enterprise customers who have grown more price-sensitive after a period of rapid capability gains without matching gains in return on investment, and competition from cheaper open-weight models developed in China, which has pushed both U.S. labs to compete on cost as well as raw performance. Randall Hunt, chief technology officer at the cloud consultancy Caylent, said corporate finance chiefs have experienced "sticker shock" from AI spending without proportional productivity gains, driving a push toward optimizing cost per completed task heading into 2027 budgets.

Anthropic said Opus 5.5 was evaluated before release by outside groups including the AI-safety research organization METR, and that the company worked with the U.S. National Institute of Standards and Technology's Center for AI Standards and Innovation on cyber and biological risk testing. Anthropic reported an 85 percent drop in self-identified "boundary circumvention" attempts compared with prior models, all rated low severity, and said the release includes watermarking required under the European Union's AI Act along with an option for customers to opt out of data retention entirely. OpenAI, for its part, positioned Sol as closing in on the reliability of Astra, the flagship model it launched earlier this month, while charging roughly half as much per token — a direct comparison OpenAI drew against Anthropic's pricing in its own announcement.

Who is affected

The immediate audience is enterprise and developer customers choosing between the two ecosystems for coding assistants, customer-support automation and data-processing pipelines, where token costs scale quickly with usage. Cloud providers that resell both companies' models, including Microsoft, Amazon and Google, stand to see usage patterns shift toward whichever option offers the best price-to-performance ratio for a given task, and the price cuts extend to consumer-facing products as well: OpenAI made Luna available to free and low-tier ChatGPT users through its desktop app, broadening access to a low-cost model beyond paying enterprise and developer customers. Software teams that build coding tools on top of either company's models are likely to see the most direct impact, since coding and agentic workflows are typically the most token-intensive and price-sensitive use case for both Sol and Opus 5.5.

What happens next

Anthropic said Sonnet 5.5 and Haiku 5.5, smaller and cheaper models in the same family, are coming in the following weeks, which would extend the price competition down the product line. OpenAI has not announced a further update to its GPT-6 family beyond Sol and Luna. Industry watchers say the more important test will come in the following earnings cycles, when both privately held Anthropic and predominantly private OpenAI provide any indication of whether repeated price cuts are being offset by usage growth, or are simply compressing margins across the industry as competition intensifies.