Anthropic launched Claude Opus 5 on Friday, describing the new system as its best-performing and most cost-effective AI model to date. The company says Opus 5 outperforms its previous model, Fable 5, on coding and knowledge-work benchmarks while offering lower pricing for customers.

Introducing Claude Opus 5. It's a thoughtful and proactive model that comes close to the frontier intelligence of Fable 5 at half the price. pic.twitter.com/GQWhcq2CQL — Claude (@claudeai) July 24, 2026

Opus 5 is priced at $5 per million input tokens and $25 per million output tokens. Anthropic said that is half the cost of Fable 5, positioning the model as a lower-cost option for businesses monitoring AI spending. The pricing model—charging separately for text sent into the model and text it generates—has become the industry standard, allowing companies to forecast costs based on usage volume.

Anthropic describes Opus 5 as a model “designed to be used every day.” It sits below Fable 5 and Mythos 5 in the company’s lineup, which now includes four entries in the Claude 5 family.

Mythos 5 remains Anthropic’s most powerful model, but access is limited to a select group of companies. Anthropic has restricted availability because of the model’s advanced cybersecurity capabilities.

Cost Pressure Shapes Model Strategy

Enterprise customers have become more cautious about AI spending, increasing pressure on model developers to deliver stronger performance at lower cost. Anthropic’s head of product management for research, Dianne Penn, said customers are looking for value rather than simply cheaper products.

“If it’s a cheaper model or a cheaper offering, but it’s not accomplishing a similar level of quality, it’s actually not useful,” she told CNBC.

Anthropic is facing pricing pressure from Microsoft, Amazon, Google, and several Chinese AI startups. OpenAI released its GPT-5.6 family earlier this month, while Google is reportedly preparing a Gemini 3.5 Pro release, although that launch has reportedly been delayed.

China’s Moonshot AI also released its open-weight Kimi K3 model last week. The lab says Kimi K3 nearly matches both Fable 5 and GPT-5.6 Sol in performance. Open-weight models like Kimi K3, which allow developers to download and run the model on their own infrastructure, have intensified competition by offering a free or low-cost alternative to API-based services like Anthropic’s.

IPO Timelines Add Pressure

The latest model launches are taking place as major AI companies prepare for potential public listings. Anthropic filed confidential paperwork for an initial public offering in June and appears on track to go public later this year.

OpenAI filed shortly afterward but may delay its IPO until 2027. The company is also facing a trade secrets lawsuit from Apple, which alleges that OpenAI used proprietary Apple information, adding complications to its path toward a public listing.

Earlier this year, the Trump administration forced Anthropic to remove Mythos 5 and Fable 5 from the market over cybersecurity concerns. Access was restored after roughly two weeks of negotiations. The episode highlighted a new dimension of AI competition: model releases are now subject not only to market forces but also to direct government intervention when agencies judge capabilities to pose potential risks.

Anthropic says it continues to work with government agencies on pre-deployment testing, including for Opus 5. The company confirmed that Opus 5 remains behind Mythos 5 in offensive cybersecurity and biology research capabilities.

Unlike Fable 5, Opus 5 does not retain user data for 30 days. Anthropic holds Fable 5 data as part of its safety monitoring efforts. The data-retention distinction may matter to customers in regulated industries, where shorter retention periods can reduce compliance burdens.

The AI model race continues as companies balance performance, cost, and regulatory requirements heading into a high-stakes year for the industry.