Anthropic launched Claude Fable 5, which secured the top rank on the Artificial Analysis Intelligence Index with a score of 64.9, placing it nearly five points ahead of the closest non-Anthropic model, GPT-5.5.
- The model uses an adaptive reasoning approach with Claude Opus 4.8 as a fallback for safety-flagged queries, which occurred in approximately 9% of tasks on benchmarks like Humanity’s Last Exam and AA-Omniscience.
- It achieved state-of-the-art results across all three agentic evaluations, including GDPval-AA, Terminal-Bench Hard, and Tau2-bench Telecom, while setting the highest score on five of the ten underlying benchmarks.
- Claude Fable 5 scored 53% on Humanity’s Last Exam, more than seven points ahead of Claude Opus 4.8 (max), and led AA-Omniscience with a score of 40 driven by higher accuracy.
- The model retains a 1M token context window but is priced at $10/$50 per 1M input/output tokens, double the cost of Claude Opus 4.8, resulting in an estimated $2.2k cost to run Humanity’s Last Exam.
The launch demonstrates Anthropic's continued lead in frontier intelligence and agentic capabilities, though the high pricing and fallback mechanisms impact overall efficiency.