Anthropic has introduced Claude Fable 5.1, a significant update to its flagship Claude family model that promises to revolutionize the way developers and researchers interact with artificial intelligence. Released on July 1, 2026, the new version focuses on improvements for code, search and tasks that require long processing hours. The most notable change is the 75% reduction in cache read cost, making the model more competitive in terms of price per completed task.
Cheaper cache redefines usability for Claude Fable 5.1
While the price per million tokens for input and output for Fable 5.1 remains the same as the previous version, $10 and $50 respectively, reducing the cache cost to $0.25 per million tokens drastically changes the cost dynamics. This means an estimated savings of 25% on typical workloads and up to 45% on highly agentic workflows. In practice, Anthropic aims to position the Fable 5.1 not only as the most powerful model, but also as an economically viable option for various daily operations.
This company strategy focuses directly on the cost per completed task, an essential criterion for software companies. Previously, many teams prioritized Opus 5 for routine code due to its lower price per token, reserving Fable 5 for only the most complex challenges. With more affordable caching, Fable 5.1 seeks to fill this gap, offering superior performance at a more efficient cost in scenarios where context is frequently reused.
Performance boost for long code and scientific research
Claude Fable 5.1 is optimized to handle projects that involve thorough analysis of application code, extensive reviews, multi-day stand-alone sessions, and scientific work such as designing experiments and interpreting dense data. Anthropic reported significant improvements in several performance metrics, especially in agentic search.
The most evident jumps in its capacity are observed in the following benchmarks:
- Terminal-Bench-Science 0.1:It reached 52.6% on Fable 5.1, a notable increase compared to 24.7% on Fable 5, 29.0% on Opus 5, and 22.4% on GPT-5.6 Sol.
- Terminal-Bench 4.0 (agentic code):Fable 5.1 achieved 55.8%, and Mythos 5.1, 60.9%, surpassing Fable 5’s 42.0% and Opus 5’s 52.3%.
- CursorBench 3.2.0:With 73.4%, Fable 5.1 was ahead of Fable 5 (70.5%), Opus 5 (70.0%), and GPT-5.6 Sol (67.2%).
- AutomationBench:Fable 5.1 recorded 31.4%, compared to 17.1% for Fable 5.
- OSWorld 2.0 (computer use):The model achieved 77.9% in partial results and 41.7% in strict results.
These numbers indicate that for low- or medium-effort tasks, the Fable 5.1 ties or surpasses its predecessor, but at a reduced cost. The model is identified in the API as “claude-fable-5-1”, and its knowledge cutoff is dated June 2026.
Customer testimonials highlight the effectiveness of the new model
Customers who gained early access to Claude Fable 5.1 reported a qualitative pattern of response that focuses on the root cause rather than superficial solutions. Manager Millennium, for example, reported that Fable 5.1 identified the origin of a rare “crash”, which occurred in approximately one in every thousand executions, a problem that engineers and other models, including Fable 5, were unable to explain for years.
Jane Street Capital stated that the model solves more code problems than Fable 5 and Opus 5, while maintaining clarity in prolonged tasks. MongoDB described creating a prototype in about three days, with the model operating for hours without the need for supervision. These experiences reinforce Fable 5.1’s ability to deliver deeper, more autonomous results.
Data Security and Privacy with Mythos 5.1 and EFS
Anthropic also released Claude Mythos 5.1, a version of the same Fable 5.1 model but with additional safeguards. Restricted to trusted access programs, Mythos 5.1 is designed for cybersecurity and life sciences applications. While Fable 5.1 can be used to discover software vulnerabilities, the company clarifies that it should not be used to develop exploits. The new security barriers in Mythos 5.1 result in approximately 60% fewer false positives in cybersecurity and an approximately 85% reduction in the refusal rate in biology and basic medicine.
An important customer demand, data retention, is addressed by the Enterprise Frontier Safeguards (EFS) system. This solution promises the equivalent of “zero retention” by storing data in the cloud controlled by the customer, not Anthropic. The EFS will be implemented in phases starting in the autumn in the northern hemisphere. Until then, eligible customers can now enjoy Fable 5.1 with the zero data retention option.
Changes to API integration rules
Anthropic reported three important changes that could impact existing API integrations:
- Forcefully using tools will now result in an error.
- Reasoning blocks can only be read by the model that generated them.
- Editing previous turns invalidates these reasoning blocks.
These changes aim to improve consistency and security in interactions with models. Fable 5.1 is available to Pro, Max, Team and Enterprise users, both on the Anthropic platform and on cloud services from Amazon Web Services, Google Cloud and Microsoft Foundry. For inference carried out in the United States only, the cost is 1.1 times the entry and exit price.
Anthropic’s strategy in the competitive artificial intelligence market
The release of Fable 5.1 comes at a strategic time for Anthropic, after Opus 5 has established itself as the “daily standard” for many code operations, mainly due to its cost-benefit. Anthropic’s response was not to reduce Fable’s list price, but rather to lower the cost of what most impacts long agent operations, the cache, and boost performance in areas where the original Fable 5 had weaknesses, such as agent science.
For the user who pays the API bill, the choice now transcends the mere comparison between “Fable or Opus?”. The decision depends on whether the task requires constant re-reading of the same context. In these cases, Fable 5.1 may be more cost-effective per delivery, even with a higher storefront price. In short, non-cache-intensive workloads, Opus 5 remains the most affordable option, following Anthropic’s own guidelines.
This launch also aligns with Anthropic’s corporate momentum, which recently reported annualized revenue exceeding US$65 billion and is exploring a large public offering. In an increasingly competitive AI market, Fable 5.1 reinforces the company’s ability to deliver robust models that support full code repositories, multifaceted searches, and a predictable invoice while meeting market demands.
