Anthropic announced Fable 5.1 and Mythos 5.1 today. The company reported notable benchmark improvements: Terminal‑Bench rose from 24.7% to 52.6%, and Mythos 5.1 reached 60.9%. For comparison, GPT‑5.6 Sol scored 22.4%.
Cost and operational changes
Anthropic kept the same pricing as Fable 5 but introduced technical changes to cut operating costs. Cache read expense fell by 75%, now $0.25, reducing overall costs by about 25%. Agent workload costs decreased by roughly 45%. Nevertheless, some tasks remain costly: one cited task cost $5.69 to run on Anthropic’s model, while GPT‑5.6 Sol executed the same task for $0.88.
Safety and access adjustments
The release also relaxed certain safety restrictions: safety false positives dropped by 60%, and defensive vulnerability analysis is now permitted. At the same time, Anthropic locked the history context to hinder model distillation and added invisible watermarks.
Market impact and competitiveness
The updated metrics widen Anthropic’s performance lead on benchmarks — Fable 5.1 is described as “uncontested” for the moment. However, market behavior suggests a different story: Fable 5 previously captured only 6% of enterprise usage because many customers preferred the cheaper Opus 5.
Analysts note that Anthropic could push performance further, but the market often rewards price‑performance rather than absolute top performance. Competitors may only need to deliver 90% of the leader’s performance at a tenth of the price to win customers. The article frames this as a “leader’s curse”: the company is advancing quickly, but the market may be unwilling to pay a sustained premium.
Conclusion
Technically, Fable 5.1 and Mythos 5.1 represent meaningful improvements — better benchmark results, lower operating costs, and relaxed safety false positives. Yet Anthropic’s launch centered on defending an intelligence premium that many customers appear reluctant to pay, highlighting a tension between top-end capability and market price sensitivity.
