OpenAI and Anthropic Ignite AI Price War With New Models

The price list shows why. Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, down from $5 and $25 on Opus 5. Cache reads fall 60 percent to $0.20 per million tokens. Anthropic says the model generates output more than 30 percent faster than Opus 5, and that the 40 percent saving measures the cost of completing a typical workload, since the model uses fewer tokens per task. The company said cheaper Claude Sonnet 5.5 and Claude Haiku 5.5 models will follow in the coming weeks, as reported by AI Stock Wire. In its announcement, Anthropic said Opus 5.5 "requires less compute to serve than Opus 5, and its pricing reflects that," according to 9to5Google's coverage. Both announcements frame the AI price war as a fight over cost per finished task, not just list prices, with each lab emphasizing how few tokens its models need.

OpenAI's answer, hours later

OpenAI did not leave the announcement unanswered. By early afternoon in New York, the company had released GPT-6 Sol and GPT-6 Luna, its next generation of mid-tier models. GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens, half of what Anthropic lists for Opus 5.5. GPT-6 Luna is priced lower still, at $0.10 and $0.50 per million tokens, aimed at high-volume routine work such as document summaries and quick questions. OpenAI kept its flagship GPT-6 Astra at the top of the ladder at $10 and $50 per million tokens, according to ZeroHedge's account of the day's launches.

Each OpenAI model costs roughly half what its GPT-5.6 predecessor did. Sol targets coding, data analysis, and multi-step reasoning, while both new models carry a context window of about 1.05 million tokens. OpenAI says cached input gets a 90 percent discount, which puts Sol's cache reads at the same $0.20 rate Anthropic charges. The AI price war is now visible on developer pricing pages, where Sol undercuts Opus 5.5 by half on both input and output tokens.

A launch day that undercut a truce

The AI price war broke out ten days after Anthropic chief executive Dario Amodei published an essay urging the industry to pace frontier development. On September 12, Amodei wrote that pacing "does not mean halting model training or technical progress," but means taking time to align and safeguard models. OpenAI chief executive Sam Altman agreed, and Elon Musk wrote that Amodei was right.

Ten days later, the two labs spent a Tuesday trading headlines. Leakers had pointed to GPT-6 Sol arriving Tuesday at $2.50 and $15 per million tokens, a forecast that proved too high, and some reports claimed Anthropic rushed Opus 5.5 out the door to beat OpenAI to the announcement, according to ZeroHedge. Musk, who had backed the slowdown call, congratulated Anthropic on the launch.

Both labs published their own benchmarks

On Anthropic's published comparison, Opus 5.5 scores 54.4 percent on FrontierCode against 53.3 percent for OpenAI's flagship GPT-6 Astra, while Astra leads on business-workflow automation and scientific research, with Anthropic noting that the GPT-6 figures are the ones OpenAI reported.

OpenAI's charts, in turn, compare Sol with last-generation Claude. The company touts 33.2 percent on AutomationBench at 27 cents a task, against 26.9 percent for Opus 5 at about eleven times the cost. Anthropic's rebuttal is that Opus 5.5 scored 40.0 percent on the same test and needs fewer tokens to finish each job. OpenAI also included a footnote claiming Fable 5.1 fell back to Opus 5 on about two in five tasks. The comparisons mix company-run evaluations and previously published results, with model versions and test settings varying, which limits how directly they can be compared.

OpenAI also said Sol made about half as many factual errors as its predecessor on an internal evaluation. Anthropic said Opus 5.5 went through extensive internal safety testing and was reviewed before release by outside organizations, including METR and Frontier Design.

Why the price cuts matter beyond the leaderboard

API prices shape the economics of the AI industry. They decide which startups can afford to build on a frontier model and how quickly developers can iterate. A single day that roughly halves mid-tier pricing puts direct pressure on open-weight models competing for token share, and it sets a new floor that developers will expect future releases to meet or beat.

For developers, the AI price war puts a number on the trade-off: Anthropic's premium model now lists at twice OpenAI's mid-tier rate for input tokens.

Related reading: Palo Alto AI Cybersecurity: Rival Models Join the Defense, on how the same two labs' models are being deployed together in enterprise security products.

The models are available now through the usual cloud channels. Claude Opus 5.5 runs on Amazon Web Services, Google Cloud, and Microsoft Azure, while OpenAI's models run on Microsoft's cloud. Neither Anthropic nor OpenAI is publicly traded, so the AI price war reaches investors only through those cloud partnerships.