Fable 5.1 Launches With Benchmark Gains but Steep Reasoning Costs

Claude Fable 5.1 is available through Perplexity Computer for Pro and Max subscribers, where Perplexity reports it ranked first in their August WANDR evaluation at a score of 0.601 and $12.76 per task, 21% higher score and 37% lower cost than Fable 5. The model offers five reasoning effort levels; independent tests by Simon Willison found that at low and medium effort it appeared to skip reasoning with costs around $0.10, while max effort produced better output at $3.30 and nearly 14 minutes of runtime. A code generation experiment tasked Fable 5.1, Fable 5, and Opus 5 with recreating Lord of the Rings landmarks in Three.js; Fable 5.1 produced the most visually detailed results but hit Anthropic's 128K output ceiling mid-file after spending 102,116 tokens on reasoning, while Fable 5 was the cheapest and fastest and the only model to complete the Bag End scene in a single call. On Code Arena: WebDev, GPT-6 Astra (Max) leads at 1,797 points, with Fable 5.1 (Max) second at 1,762 and Opus 5 (Max) third at 1,688.