Anthropic’s New Claude Is Smarter, Cheaper and Says “No” Less

Anthropic’s New Claude Is Smarter, Cheaper and Says “No” Less

Imagine asking an AI to help you investigate a complicated problem, write code, analyze research, and keep working for hours without constantly stopping to say it cannot help.

That is the direction Anthropic is taking with its newest models.

On September 1, Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1, two versions of the same underlying AI model designed for demanding coding, research, and knowledge-work tasks. But this is not just another “new model, slightly better benchmarks” release.

Anthropic is also trying to fix two things users care about deeply: cost and unnecessary restrictions.

Meet Fable 5.1 and Mythos 5.1

The easiest way to understand the two models is this:

Fable 5.1 is the version designed for broader use. Mythos 5.1 is the more unrestricted version for carefully vetted organizations working in areas such as cybersecurity and life sciences.

They share the same underlying model, but Anthropic applies different safeguards depending on who is using it and what they are allowed to do.

Fable 5.1 is available through Anthropic's API and cloud platforms, while Mythos remains restricted to trusted-access programs. 

And Anthropic is making a big performance claim.

The company says Fable 5.1 sets new highs across several coding, reasoning, and knowledge-work evaluations, including Terminal-Bench 4.0 and Humanity's Last Exam. It also reports major improvements on a scientific-research benchmark, with Fable 5.1 scoring 52.6% on Terminal-Bench-Science 0.1 compared with 24.7% for Fable 5.

But the numbers are not the only interesting part.

The AI Just Got Cheaper to Run

Anthropic has also changed the economics of using Fable 5.1.

The company's cache-read pricing has been cut by 75%, which can make typical workloads around 25% cheaper and highly agentic workloads up to 45% cheaper, according to Anthropic. Base input and output pricing remains unchanged.

Why does that matter?

Because AI is increasingly moving beyond simple questions and answers.

An AI agent might read a large amount of information, work through several steps, use tools, revisit previous information, and continue working for hours.

That can get expensive quickly.

Making those long-running tasks cheaper could make AI agents much more practical for businesses.

And Fable 5.1 Does Not Say “No” as Often

Here is another major change.

Anthropic has refined Fable's safety systems to reduce false-positive situations where the AI blocks a legitimate request because it mistakenly thinks the request is dangerous.

Anthropic says Fable 5.1's biology safeguards intervene on benign requests 85% less often than the safeguards introduced with Fable 5.

The model can also identify software vulnerabilities in source code, although Anthropic still blocks penetration testing, exploit generation, and certain other high-risk cybersecurity activities.

So, the goal is not to remove safety.

It is to make the safety system more precise.

Privacy Is Getting an Upgrade Too

Anthropic is also pushing harder on enterprise privacy.

Its new Enterprise Frontier Safeguards are designed to give organizations more control over security monitoring and allow monitoring data to remain within infrastructure controlled by the customer. The company says the system is expected to roll out in the fall.

Anthropic also reiterated that it does not train its models on enterprise data without explicit permission.

That could become increasingly important as companies start giving AI agents access to internal documents, software, research, and business systems.

But There Is a Bigger Story

Anthropic is not simply trying to build a chatbot that answers questions better.

It is building AI that can stay on a problem longer.

The company highlighted examples involving software debugging, scientific research, and long-running machine-learning experiments. One customer reported a 38-hour machine-learning run in which Fable 5.1 continued evaluating results, launched additional experiments, and returned with findings and suggested next steps.

That points toward a different future for AI.

Instead of:

“Ask AI → get answer.”

The workflow becomes:

“Give AI a goal → let it investigate → let it work → come back with the result.”

And that is where things get really interesting.

Because the more capable these systems become, the more important it becomes to make them affordable, reliable, private, and controllable at the same time.

Fable 5.1 is Anthropic's latest attempt to put all four pieces together.

The AI race is no longer just about who has the smartest model. It is increasingly about who can make that intelligence useful enough to leave running all day.