Anthropic Rolls Out Claude Fable 5.1 With Sharper Technical Reasoning and Cut-Rate API Pricing
Advancements in artificial intelligence models usually come at the expense of high compute costs. This time, Anthropic, an AI research safety lab, has decided to buck the trend and announced its new release: Claude Fable 5.1 and Mythos 5.1, its enterprise counterpart. These latest versions feature improvements in advanced coding and long-horizon research, along with lower API token costs for developers.
Responding to continuous criticism about overprotective safety filters, the latest release reduces false positive refusals by around 60% in software engineering applications. Now, when using Claude Code, developers can analyze codebases for security vulnerabilities without shutting down their systems due to the wrong code.
Although the model is allowed to find vulnerabilities in the software, Anthropic still has strong control over it and won’t allow Claude to automatically generate any exploit code for it. Besides, Fable 5.1 also features enhanced problem-solving with a focus on finding the root of the problems in the codebase.
Financially, it makes running autonomous, agent-based loops cheaper. Although input and output token costs stayed the same, Anthropic reduced prompt cache read cost by 75%, making it $0.25 per million tokens instead of $1.00. As far as workflows that require frequent reinsertion of large code repositories or vast amounts of documentation go, the average workload will experience around a 25% to 45% reduction in operational costs.
The release also signals the adoption of a dual release model, designed in compliance with regulations. Even though Fable 5.1 is freely accessible on all commercial cloud services, its counterpart Mythos 5.1 stays within trusted access programs, including a Life Sciences Verification Program for biological research purposes.
By combining increased reasoning capabilities with substantial caching discounts and reduced operational restrictions, Anthropic aims at solving critical issues faced by developers. Reducing the cost and bureaucratic burden associated with constant multi-turn processing makes it possible to develop sophisticated and lengthy AI agents economically.
