Anthropic has introduced Fable 5.1 and Mythos 5.1, signaling how it intends to address pricing concerns for enterprise customers — even as the release underscores that enterprises do not always need the most powerful model for their workloads.

On Sept. 1, the AI lab revealed that Fable 5.1 and Mythos 5.1 are the same models, except that Mythos 5.1 is accessible only to authorized users as part of Anthropic's Project Glasswing cybersecurity initiative and includes distinct levels of safeguards.

Alongside increased capabilities in advanced coding and knowledge work, and improved research skills in scientific fields, Fable 5.1 costs 25% less than Fable 5 for typical workloads because it reduces cache read pricing. A cache read occurs when the model reads inputs that have already been processed or stored — a cost that accumulates for enterprises running repetitive or high-volume workloads against the same prompts and documents. Fable 5.1's cache read pricing drops from $1 per million tokens to $0.25 per million tokens. Standard input pricing remains $10 per million input tokens and $50 per million output tokens — still significantly higher than OpenAI's GPT-5.6 Sol, priced at $5 per million input tokens and $30 per million output tokens.

An Escalating Price War

Fable 5.1's price cut reveals the vendor's strategy for dealing with the escalating price war in the AI market. Alongside Anthropic, OpenAI, Google and Microsoft are competitively reducing prices as the cost of AI becomes harder to contain, while enterprises try to determine the ROI and use cases for frontier models.

"We are at a juncture where token consumption is a big concern," said Lian Jye Su, an analyst at Omdia, a division of Informa TechTarget.

Su added that while vendors have begun to reduce prices, enterprises have started finding their own ways to address the pricing issue.

"People are using a lot more model routing nowadays compared to just relying on a single vendor," Su said. "The type of processes that you run will dictate what type of models you select at the end of the day. If you are someone that really needs that frontier capability, you will still go with the most complex models."

Pricing and Anthropic's IPO

Anthropic's move to reduce cached read prices for Fable 5.1 — while also claiming the model is the same as Mythos 5, its most powerful and restrictive model, with a separate set of safeguards — might be intended to further its IPO ambitions. Anthropic filed for an IPO on June 1 and is expected to go public this fall. Competitive pricing against rivals such as OpenAI's GPT-5.6 Sol gives the vendor a way to appeal to cost-conscious enterprise buyers ahead of that listing.

"It needs to have this solid recurring revenue to ensure that its valuation continues to stay on top," Su said.

Not for All Enterprises

Lower pricing, combined with Fable 5.1's improved performance, makes the model attractive to enterprises. However, enterprises should still assess whether the model is truly needed for their specific workloads.

"This is definitely an announcement for companies that are looking at the most frontier capability that really care about differentiating themselves from the rest," Su said. He added that the model is not for back-office operations or day-to-day tasks, nor for enterprises that need an on-premises option.

"New models nowadays focus very strongly on agentic AI vulnerability testing or advanced challenges," Su said. "Most of the stuff that we do with AI nowadays is not necessarily the most complicated tasks."

Strengthened Safety Measures

Even if enterprises do not need Fable 5.1, the model shows how Anthropic continues to strengthen its safety measures. The vendor revealed that until its Enterprise Frontier Safeguards are available, customers can use Fable 5.1 with zero data retention.

Anthropic introduced Enterprise Frontier Safeguards on Sept. 1 as a new security standard that combines the privacy of not retaining enterprise data with the security of detecting misuse. The vendor's newest safeguards are also better at reducing false positives — instances when an AI model flags a pattern or object that is not actually present, or flags unimportant content.

"Anthropic has always been strongly against AI that is left ungoverned," Su said. "It's that continual effort to ensure it stays on top of all the critical concerns in the market."