Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1, framing them as the world’s most advanced models for coding and knowledge work. The release coincided with a 75% price cut for cache reads, intended to benefit long-running sessions and context-heavy users. However, Artificial Analysis Intelligence Index data revealed a 1.7x increase in output token usage, resulting in a 20% net increase in per-task costs.
The models were evaluated across multiple benchmarks, including Terminal-Bench-Science, SWE-family evals, HLE, FrontierCode, and Artificial Analysis. Initial community reactions highlighted strong coding and planning abilities alongside concerns about rate limits and safeguards. Analysis suggested the underlying weights were the same, differing primarily in safety and routing behavior, rather than a fundamentally new base model.
Key technical details included a cache read price reduction from $1.00 to $0.25 per million tokens, and a maximum intelligence score of 66, as measured by Artificial Analysis. Comparisons with Opus 5 showed Fable 5.1 achieving similar performance on some measures, particularly in agentic knowledge work. A fallback mechanism to Claude Opus 4.8 or 5 was observed, accounting for approximately 4% of output tokens across the Intelligence Index.
Cost-performance analysis revealed a trade-off: Fable 5.1’s lower max score (66) resulted in a 20% higher cost per task compared to Fable 5.1’s max score (65). Sol remained the clear winner on intelligence-per-dollar and intelligence-per-token. The launch included new enterprise-oriented controls, specifically Enterprise Frontier Safeguards (EFS), positioned as ZDR++ for agent observability. Zero-data-retention support was also highlighted as a key adoption unlock.
Source: https://www.latent.space/p/ainews-claude-fablemythos-51-new
