Sol's 27-cent benchmark task undercuts Opus 5 by more than eleven times

Anthropic debuted Claude Opus 5.5, which it says performs at the level of its flagship Fable 5.1 model but costs around 40% less to run than Opus 5, which came out in July.

OpenAI released GPT-6 Sol and GPT-6 Luna, offshoots of its flagship GPT-6 Astra model, and cut the API cost by 50% compared with the pricing on a promotion it is currently running for GPT-5.6.

OpenAI released GPT-6 Sol and GPT-6 Luna on Tuesday, minutes after Anthropic launched Claude Opus 5.5.

Opus 5.5 underwent Anthropic's most rigorous alignment testing to date, was evaluated by outside groups, and launches with the same tier of cybersecurity, biology and frontier-AI-development safeguards previously reserved for Fable 5.1.

CNBC described Claude Opus 5.5 as a more token-efficient version that will cost around 40% less to run than Anthropic's Opus 5.

"One of the things we're continuing to innovate on is how to make that thinking, how to make the answering more efficient, so it uses less tokens depending on your effort setting," Dianne Penn, head of product management, research and labs at Anthropic, told CNBC in an interview.

Anthropic said the model is its most efficient, and it is "passing these efficiency savings on to our customers in the form of price cuts and rate limit increases."

OpenAI attributes the cost reductions to "improvements in caching and inference" and says it is "passing those savings directly onto users and customers."

"OpenAI and Anthropic are engaged in a price war that is driving down the price of AI and therefore driving down their ability to profit from it and grow the price of models," Ara Kharazian, lead economist at Ramp, told Fortune.

"AI bulls assume that there will be highly performant models that provide more and more value, and therefore they should be more expensive," Kharazian said. "But that is not how normal technology makes it to market."

The labs are facing pressure from companies offering cheaper, open-weight models, including Chinese firms Alibaba, Moonshot AI and DeepSeek.

Opus 5.5 is the first release in a new family of models, with Sonnet 5.5 and Haiku 5.5 expected "over the coming weeks," Anthropic said.

OpenAI's release offers lower API pricing than Anthropic's, based on each company's announcement.

Sol, a tier below Astra, is meant for more complex workloads like coding, while Luna is geared toward "high-volume tasks" like extracting information or summarizing documents.

Neither Sol nor Luna is in the plain ChatGPT app yet, and OpenAI says it is rolling both out gradually throughout the day to keep the service stable.

"CFOs have seen some of the sticker shock, and they haven't seen some of the gains that were promised in the initial investments," said Randall Hunt, CTO at the AI consulting company Caylent. "So they're coming back to their planning for 2027 and beyond, and saying, 'How can we optimize our costs here?'"

Caylent calculates how long it takes employees to complete tasks with AI tools and then calculates "cost per task," an approach OpenAI encouraged in a mid-July 2026 blog post.

Luna's input price fell by half while its output price fell about 58%.

A Luna job that reads one million tokens and writes one million tokens now costs $0.60, against $1.40 at the previous rates.

Sol's new prices are exactly half the old rates on both the input and output legs, so an identical token volume costs half what it did at the previous rates.

Reasoning effort is a dial OpenAI added: turn it up and the model spends more time and computing power double-checking itself before answering, which tends to raise accuracy and cost together.

On AutomationBench, a benchmark built by Zapier that tests whether an AI agent can carry out a full business workflow using 47 tools spanning sales, marketing, operations, support, finance and HR, GPT-6 Sol at its highest reasoning setting scored 33.2% for $0.27 per task.

Claude Opus 5 at its own top setting scored 26.9% on AutomationBench, but cost more than 11 times as much per task, according to OpenAI's own numbers.

The releases from both companies are not frontier models, in that they are not a major step up in capabilities.

At more than 11 times Sol's $0.27, Opus 5 cost more than $2.97 per AutomationBench task, about 11 cents for each point of pass rate against Sol's 0.8 cents.