Anthropic has officially unveiled Claude Fable 5.1 and a limited version of Mythos 5.1, calling them its cutting-edge solutions for coding and knowledge work. The release took place on Tuesday, and alongside updated algorithms, the company introduced a fundamentally new security measure aimed at protecting the model's internal "reasoning" from unauthorized copying by competitors.
The key change concerns API accounts created after August 31. Developers can no longer manually edit the conversation history with Claude while preserving the transcript of its previous reasoning. This closes a loophole that was actively exploited for distillation—the process of training cheaper models on the responses of more expensive, higher-quality counterparts.
Pricing and performance: a bet on savings
The cost of base tiers remains unchanged: $10 per million input tokens and $50 per million output tokens. However, the economic model has shifted significantly. Cache reads have dropped by 75%, down to $0.25 per million operations. According to Anthropic's estimates, typical tasks now cost 25% less, while complex scenarios for AI agents offer up to 45% savings.
In its own benchmarks, the company reports impressive progress. In the Terminal-Bench-Science 0.1 test, which evaluates "scientific" agent work, Fable 5.1 scored 52.6%, whereas the previous version, Fable 5, managed only 24.7%. In coding, the improvement is more modest, but the result is still compelling: 55.8% on Terminal-Bench 4.0 versus 52.3% for Opus 5 and 37.3% for OpenAI's GPT-5.6 Sol. The model's knowledge base now covers data up to June 2026—five months more than its predecessor.
The war on distillation: numbers and accusations
The issue of copying has long been a headache for American labs. Back in February, Anthropic cited specific data: over 16 million distillation-related conversations were initiated through roughly 24,000 fake accounts. The company directly named three Chinese developers—DeepSeek, Moonshot AI, and MiniMax. Moonshot accounted for 3.4 million requests.
In July, White House science advisor Michael Kratsios went even further, accusing Moonshot of copying Anthropic's flagship model to build its own Kimi K3. Moonshot has not publicly responded to these allegations.
Practical implications for the market
The new rules apply only to accounts created after August 31. Users of Claude Code, Cowork, and Claude.ai will not notice any changes, and older API keys will continue to operate as before. Anthropic promises that in upcoming versions, all accounts will undergo verification. Fable 5 will remain available until at least June 2027, so there is no need to rush the migration. Support for Fable 5.1 is guaranteed until September 2027.
My take: This move is not just a technical update but a signal that the open phase of confrontation in the AI industry has begun. Anthropic is not merely patching a vulnerability; it is demonstrating a willingness to fiercely protect its research investments. For the market, this means the arms race between labs is shifting from "who is faster" to "who is better protected." Distillation as a method will not disappear, but it will become significantly more expensive and complex, which in the long run could slow the emergence of cheap imitation models.