The "arms race" of AI large models is now fully extending toward extreme cost-effectiveness. As the key milestones for going public approach, Anthropic has officially launched its new flagship large model Fable5.1, and simultaneously introduced Mythos5.1 to users with Mythos access permissions. As the most powerful core in Anthropic's product lineup, this model maintains its strong performance while significantly reducing actual usage costs.

Regarding core model performance, Fable5.1 has comprehensively surpassed its predecessor, Fable5, as well as competing products in multiple benchmark tests, once again pushing the boundaries of industry large language models. The biggest upgrade in the new version is the enhanced autonomous operation capabilities of "agents," allowing it to handle complex tasks that are long-term and involve many steps more smoothly. Whether it's long-term programming development, in-depth academic research, or complex knowledge work, it can independently verify results and dynamically adjust priorities during extended operations, effectively avoiding deviations or superficial outcomes.

image.png

In addition to achieving significant performance improvements, the most notable aspect for the industry is the substantial reduction in usage costs. According to official data from Anthropic, the expected overall cost of processing typical token-based workloads with Fable5.1 has decreased by about 25% compared to the previous generation. This is mainly due to a significant reduction in model cache reading costs, which drastically lowers system expenses when reusing previously processed and stored input content.

image.png

In high-intensity complex tasks involving agents, the cost savings can reach up to approximately 45%. This means that users can now run Fable5.1 at a "moderate or low effort" level, achieving high-quality results comparable to, or even better than, those of the previous generation running at a "high effort" level, with lower financial investment.