Anthropic's latest AI model, Claude Fable 5, has achieved the highest score in the Artificial Analysis Intelligence Index, with 64.9 points, surpassing its closest competitor, GPT-5.5, by five points. The model's performance gain over its predecessor, Opus 4.8, is 5.7 percent across multiple benchmarks, according to the Artificial Analysis Intelligence Index. However, this improvement comes at a significant cost, with token prices doubling compared to Opus 4.8.
A full benchmark run of the Artificial Analysis Intelligence Index now costs nearly $10,000 with Fable 5, compared to $4,970 for Opus 4.8. The model's token pricing is $10 and $50 per million input and output tokens, double the $5 and $25 rates of Opus 4.8. Anthropic noted that the performance gain is modest but tangible, and the price increase is steep, raising questions about the cost-effectiveness of the upgrade.
The model's high cost is partly due to enhanced safety filters for queries involving cybersecurity, biology, chemistry, and model distillation. These filters reroute requests to Opus 4.8 when triggered, increasing billing costs. During the Intelligence Index evaluation, about eight percent of tasks involved fallback routing, mainly on science questions from GPQA, AA-Omniscience, and Humanity's Last Exam. On the Humanity's Last Exam test alone, the fallback rate reached nine percent.
Source: thedecoder
