Anthropic unveils 2 new versions of Claude, and they achieve remarkable feats

Anthropic has just unveiled two new artificial intelligence models for Claude. As the back-to-school season begins, the OpenAI rival startup has revealed Claude Fable 5.1 and Claude Mythos 5.1. With these two new AIs, the California-based company promises solutions that are less expensive, more capable in programming, and equipped with more precise cybersecurity safeguards.The two models share the same technical architecture, but differ in the level of security of their safeguards, which are intended to prevent malicious uses at a time when AI-assisted cyberattacks are multiplying.

The main advantage of both updates is price. When AI reuses information it has already analyzed, it now costs 75% less. For Anthropic customer companies, the bill drops by around 25% for typical use, and can fall by as much as 45% for the most intensive uses, such as automated programming.

Also read: Hacking of Claude – Anthropic warns its customers of a cyberattack, 6 viruses targeting your PC

The achievements of the new Claude models

That’s not all. Anthropic also says it has improved the performance of its AI models. The American startup claims that Fable 5.1 outperforms its predecessor Fable 5, as well as its own Opus 5 model and OpenAI’s GPT-5.6 Sol. On the AutomationBench benchmark, which measures the ability to automate professional tasks, Fable 5.1 scores 31.4%, compared with 17.1% for Fable 5. That is a notable improvement.

According to Anthropic, the models have achieveda series of remarkable feats. First, Mythos 5.1 solved an extremely rare bug that had eluded engineers for nearly five years in the system of investment firm Millennium. The model disassembled an external software library to trace the origin of the bug.

In addition, Anthropic highlights the scientific achievements of its AI systems. Mythos 5.1 succeeded in creating, in a virtual laboratory, proteins capable of binding very effectively to their target, an achievement typically considered complex in medical research. The AI achieved a 50% success rate in protein synthesis, well above the 15% of conventional methods. Meanwhile, Fable 5.1 distinguished itself in astronomy. It used old radar data from the American Magellan probe, launched in the 1990s, to draw an unprecedented map of part of Venus’s surface

Fewer false positives

On the security front, Anthropic announces a 60% reduction in false positives. These are instances in which the system blocked legitimate requests out of excessive caution. In such cases, the AI firmly refuses to answer a request that is in no way criminal or malicious. These situations should therefore become rarer when you converse with the new Claude.

Furthermore, Fable 5.1 is now authorized to identify software vulnerabilities for defensive purposes. Nevertheless, the AI remains unable to generate exploit code for the vulnerabilities it identifies. Medical and biological questions also benefit from a similar improvement. The AI now blocks harmless requests 85% less often.

Note that Anthropic is also launching a new system called Enterprise Frontier Safeguards (EFS).It allows client companies to keep all their data on their own servers rather than sending it to Anthropic, while still protecting it against malicious use. This feature will roll out gradually this fall and is aimed primarily at organizations concerned about protecting their data.

Gaining Ground on OpenAI

With these two new Claude models, Anthropic is seeking to gain ground on its main rival, OpenAI. The competitor was due to launch a new flagship model, called Astra, as early as the beginning of summer. The project was put on hold after the cyberattack on the open-source platform Hugging Face, which was hacked by an army of rogue ChatGPTs. OpenAI has vowed to unveil Astra as soon as possible. For investor Gavin Baker, head of Atreides Management, the fact that “releasing Fable 5.1 before Astra is one hell of a masterstroke”.

Releasing Fable 5.1 before Astra is quite a flex.

Makes me think Fable 5.2 is ready to go.

So much gamesmanship between Anthropic and OpenAI right now.

And I am very much looking forward to the next version of Grok.

Exciting times.

— Gavin Baker (@GavinSBaker) September 1, 2026

The announcement comes as Anthropic is expected to make its debut on Wall Street.The start-up would have confidentially filed its IPO application in June 2026, with a view to entering the financial markets by the end of September.

👉🏻 Follow the latest tech news in real time: add 01net to your sources on Google, and subscribe to our WhatsApp channel.