Anthropic has landed another blow against its competitors, unveiling its new flagship models on Tuesday — Claude Fable 5.1 and a limited version of Mythos 5.1. The company states outright that these are its most advanced solutions for coding and knowledge work. The new models are already available across all major cloud platforms, including Amazon Bedrock, Google Cloud, and Microsoft Foundry.
The cost of base tiers remains unchanged: $10 per million input tokens and $50 per million output tokens. However, the real savings for developers are enormous. Cache reads have dropped by 75% — to $0.25 per million operations. According to Anthropic, typical tasks now cost 25% less, while complex scenarios for AI agents are as much as 45% more economical compared with the previous version.
Breakthrough metrics and an expanded knowledge base
Internal benchmarks show an impressive leap in performance. The "scientific" work of AI agents stands out in particular: on the Terminal-Bench-Science 0.1 test, the Fable 5.1 model scored 52.6%, while its predecessor Fable 5 managed only 24.7%. In business process automation tasks, results nearly doubled, reaching 31.4% on AutomationBench. In coding, Fable 5.1 also outpaces rivals: 55.8% on Terminal-Bench 4.0 versus 52.3% for Opus 5 and 37.3% for OpenAI's GPT-5.6 Sol.
The expanded knowledge base is also noteworthy: the model now operates with data up to June 2026, which is five months "fresher" than Fable 5.
The key innovation — protection against copying
The main new feature of Fable 5.1 is not just speed or accuracy, but a fundamentally new level of security. Anthropic has closed the loophole through which competitors could steal the model's "thinking." Now, new API accounts created after August 31 are physically unable to manually edit Claude's previous context in a multi-year conversation while retaining a transcript of its earlier reasoning.
This measure targets distillation — the process in which a cheaper model is trained on the responses of a more expensive one. The threat is not hypothetical: in February, Anthropic cited concrete data, tracking over 16 million conversations linked to distillation through roughly 24,000 fake accounts. Three Chinese labs were named directly: DeepSeek, Moonshot AI, and MiniMax. Moonshot alone accounted for 3.4 million requests.
The restriction applies exclusively to new accounts, so users of Claude Code, Cowork, and Claude.ai will notice no changes. Old records do not fall under the new rules, and support for Fable 5 will last at least until June 2027, giving developers ample time for a smooth transition.
My take: This is a strategically important move that goes far beyond a routine model update. Anthropic is not just improving its product — it is erecting a protective barrier around its core intellectual property. At a time when Chinese labs are aggressively building strength through distillation of Western models, such measures are becoming not a luxury but a necessary condition for survival in the AI market. The only question is how quickly competitors will find new workarounds.