Claude Opus 5 undercuts Fable 5 on price

Anthropic’s Claude Opus 5 beats Fable 5 on coding and knowledge tasks while cutting token prices in half.

Anthropic just pushed a sharper pricing message into the AI model race. On Friday, the company said Claude Opus 5 outperforms Claude Fable 5 on coding and knowledge work benchmarks, while costing half as much to use.

The new model is priced at $5 per million input tokens and $25 per million output tokens, and Anthropic says it is meant for everyday use. That matters because enterprise buyers have become far more selective about which models they test, deploy, and keep in production.

Price is now part of the product story

Anthropic has been under pressure to show that bigger models can still make financial sense for customers. Dianne Penn, Anthropic’s head of product management for research, told CNBC that enterprises want value, and that a cheaper model only matters if it keeps quality high.

That framing tells you where the market is right now. Buyers are no longer impressed by raw benchmark wins alone. They want a model that can justify its bill when it is used every day by developers, analysts, and internal support teams.

Opus 5 tries to answer that directly. Anthropic says it is its best-performing and most cost-effective model across a number of industry benchmarks, and it is positioned as the model people can actually leave on for routine work instead of reserving for special cases.

• Opus 5 costs $5 per million input tokens.
• Opus 5 costs $25 per million output tokens.
• Anthropic says that is half the price of Fable 5.
• The company says Opus 5 is designed for everyday use.

Anthropic is splitting capability by risk

Anthropic is also drawing a clearer line between general productivity and high-risk technical work. The company said Opus 5 is not state-of-the-art for “risky, dual-use capabilities,” including cybersecurity.

That distinction matters because Anthropic has already lived through the consequences of shipping a model that was too powerful in the wrong hands. In April, the company released Claude Mythos Preview to a limited group of users, and Anthropic said it was extremely capable in cybersecurity, including finding and exploiting software vulnerabilities.

In June, Anthropic released updated versions of that line and then briefly disabled access after a government export control directive citing national security authorities. Access returned after roughly two weeks of negotiations. That episode explains why the company is now careful about how it talks about capability.

“Enterprises, in our feedback and with our customer base, are looking for value,” Dianne Penn told CNBC. “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.”

The quote is the cleanest read on Anthropic’s strategy. The company is not trying to win every benchmark category with one model. It is trying to sell the right model for the right job, while keeping the more sensitive capabilities in a separate tier.

How Opus 5 compares with Anthropic’s other models

Anthropic’s own framing makes the lineup easier to read. Opus 5 is the everyday workhorse. Fable 5 was the previous safe general-use model. Mythos 5 remains the sharper tool for advanced security and research tasks, even if it is not the one Anthropic wants most customers touching.

That split also reflects the broader pressure across AI vendors. Anthropic and OpenAI have both been trying to persuade customers that model spend can be justified with real business output, while rivals like Microsoft, Amazon, Google Cloud, and a wave of Chinese startups keep pushing lower-cost alternatives.

• Opus 5 beats Fable 5 on coding evaluations.
• Opus 5 beats Fable 5 on knowledge work evaluations.
• Opus 5 is still behind Mythos 5 in offensive cybersecurity and biology research.
• Anthropic says it is collaborating with government agencies before deployment and testing.

For buyers, the comparison is simple: if you need a model for coding help, document work, or internal assistants, Opus 5 looks like the value play. If your use case touches security research or biology-related analysis, Anthropic is telling you that the more sensitive model tier remains separate.

For Anthropic, that separation is probably the point. It lets the company sell one model as a practical default while keeping stricter controls around the models that raise more regulatory and safety questions.

The real test is whether enterprises switch

The launch matters because enterprise AI budgets are getting more disciplined, not less. Teams that once tried every new model are now asking whether a newer release saves time, reduces cost, or improves output enough to justify a migration.

If Opus 5 can deliver the quality Anthropic claims at half the cost of Fable 5, it could become the model teams pick by default for everyday work. If the benchmark gains do not show up in real workflows, the lower price will not be enough.

That is the part worth watching over the next few quarters: whether Anthropic can turn better benchmark math into sticky enterprise usage. If it can, the company will have a strong answer to the current buyer mood. If it cannot, the market will keep treating cheaper AI as a nice headline rather than a purchasing decision.