Anthropic has officially unveiled Claude Fable 5.1 , the latest version of its flagship model, along with its limited-edition variant, Mythos 5.1. The launch, which took place on September 1st, comes three months after the release of Fable 5 and at a time when the company is looking to solidify its position against competitors like OpenAI and Google.
Both models share the same core architecture but differ in their level of safeguards. While Fable 5.1 is generally available through its API and major cloud platforms, Mythos 5.1 is reserved for verified-access programs geared toward sensitive cybersecurity and life science tasks. According to Anthropic, these are their most advanced models to date for programming, knowledge work, and scientific research.
Prices and savings
The base price of Fable 5.1 remains at $10 per million tokens for entry and $50 per million for exit, the same as its predecessor. However, the major change is in cache reads, which have been reduced from $1 to $0,25 per million tokens, a 75% decrease . This reduction has a direct impact on the overall cost, as savings of around 25% are seen in typical tasks, and can reach up to 45% in highly agent-driven workloads.
For startups and companies that use the API extensively, this change represents a welcome relief on their monthly bill . The key is that the models constantly reread the same context, and with the new pricing, this process becomes much more economical. Anthropic has confirmed that the savings are particularly noticeable in long-term programming projects and for agents that handle large volumes of data.
Performance and benchmarks
In terms of performance, Fable 5.1 sets new records in several benchmarks. In Terminal-Bench 4.0, which measures agentic coding on the terminal, it achieves 55,8%, surpassing Opus 5's 52,3% and OpenAI's GPT-5.6 Sol's 37,3%. In scientific research, the model achieves 52,6% in Terminal-Bench-Science 0.1, more than double Fable 5's 24,7%. It also improves in multidisciplinary reasoning, scoring 65% in Humanity's Last Exam when using tools.
These figures, published by Anthropic, have been reported by outlets such as Decrypt and VentureBeat. Furthermore, the Artificial Analysis Intelligence Index, an independent ranking, places Fable 5.1 in first place with 66 points, ahead of Opus 5 (63) and GPT-5.6 Sol (61). The company also highlighted that the model can match or outperform Fable 5 even with low to medium effort levels, allowing for even greater cost savings.
Privacy and Enterprise Frontier Safeguards
One of the most anticipated new features is Enterprise Frontier Safeguards (EFS), a system that allows companies to run Anthropic's models on their own cloud infrastructure, without data ever leaving their network. This represents a radical change from the previous 30-day retention policy, which was a barrier for regulated sectors such as healthcare, finance, and government.
With EFS, monitoring data is stored in the customer's account, under their own encryption keys and access policies. Anthropic maintains automated misuse detection systems, but alerts are sent to the customer for review. Rollout will begin in the fall of 2026, and until then, eligible customers can use Fable 5.1 with zero data retention . According to CNBC, the company worked "hundreds of hours" with customers to design this solution.
Security and beta features
Fable 5.1 incorporates two beta features : effort adjustment during conversation, which allows the user to change the model's intensity level without restarting the session, and content provenance tracking, which tags outputs so downstream systems can verify their origin. This latter feature is especially useful in regulated sectors, where traceability is a requirement.
In terms of security, Anthropic has improved its classifiers, reducing false positives . This allows users to work more efficiently on cybersecurity and biology tasks. The company has also published a system card that assesses the risks of Mythos 5.1, classifying it as 'low risk' in automated AI development, although it acknowledges a slight tendency towards misaligned behavior compared to Opus 5.
Scientific investigation
Anthropic took advantage of the launch to present three scientific findings generated by the models prior to their release. These include a high-resolution map of Venus, reconstructed from radar images from NASA's Magellan mission, and a custom GPU optimization that accelerates deep learning models by up to 2,5 times. They have also successfully designed molecules with high binding affinity for future drugs, achieving a success rate of nearly 50%.
These results demonstrate the potential of Fable 5.1 and Mythos 5.1 in complex scientific tasks, an area in which Anthropic aims to differentiate itself from its competitors. The company has released the Venus map under a Creative Commons license for future space missions.
Competitive context
The launch of Fable 5.1 comes at a time of intense competition in the AI ​​market. OpenAI is planning to release GPT-6, and Google continues to improve its Gemini family. Anthropic, for its part, seeks to consolidate its position with a model that offers a better balance between performance, cost, and privacy. The company has also signed the EU AI Act's Code of Practice, which entails including an invisible watermark on the generated content.
For European companies, this watermark is a key factor to consider, as it allows them to identify whether a text has been generated by AI. Anthropic has developed a detection API that will be available to regulators and media outlets. Although some experts criticize the approach as being broader than required by law, the measure strengthens transparency.
In short, Fable 5.1 arrives with significant performance improvements, reduced cache read costs, and a clear commitment to enterprise privacy with Enterprise Frontier Safeguards. This model positions itself as Anthropic's most advanced, surpassing Opus 5 in most benchmarks, and offers businesses the ability to maintain control over their data. With a phased rollout of EFS and the inclusion of the EU AI Act watermark, the company demonstrates its responsiveness to market demands and regulations. For startups and enterprises seeking a balance between capacity and cost, Fable 5.1 is undoubtedly an option worth considering.