Anthropic has introduced Claude Fable 5, a model that reportedly surpasses all previously released public models across nearly all benchmarks, including software engineering, knowledge work, vision, and scientific research, positioning it as the most advanced AI model to date.
Simultaneously, the company launched Mythos 5, a frontier model designed for security defense, alongside anti-distillation safety restrictions, offering a free trial window from June 10 to 22, with pricing set at $10 per million input tokens and $50 per million output tokens.
Shortly after release, both models were globally withdrawn following U.S. export control and security-related mandates, an event that directly impedes overseas access to advanced foreign models.
In the near term, this is expected to redirect demand back to domestic large language models, while over the medium to long term, it is likely to further stimulate demand growth for local computing power, self-developed models, AI privatization, and sovereign AI infrastructure development.
Key Points from the Analysis
Anthropic's release of Claude Fable 5 is described as its most powerful model ever, leading in almost all AI performance evaluations.
In software engineering, Fable 5 completed the full migration and compression of a 50-million-line Ruby codebase within a single day, a task that would take human engineers over two months.
It achieved the top score on the Cognition FrontierCode evaluation with only moderate effort.
For knowledge work, Fable 5 secured the highest score among all models on the Hebbia Finance Benchmark.
It has set a new state-of-the-art standard for visual tasks, capable of extracting precise numerical data from scientific charts and reconstructing web application source code from screenshots alone.
Fable 5 can also process millions of tokens in long-duration tasks, demonstrating a threefold performance improvement over Opus 4.8 in the game "Slay the Spire."
Claude Mythos 5 was launched concurrently, deployed via Project Glasswing for cybersecurity defenders and infrastructure providers, showing remarkable capabilities in scientific hypothesis generation and drug design.
Utilizing Mythos 5, Anthropic increased the efficiency of certain stages in the drug design process by approximately tenfold.
Mythos 5 is noted as the first model capable of consistently generating novel scientific hypotheses; in blind tests against Opus-level models, scientists showed a roughly 80% preference for hypotheses from Mythos.
The pricing for both models is set at $10 per million input tokens and $50 per million output tokens, which is less than half the cost of the Mythos Preview but twice as expensive as Opus 4.8.
Subscribers were granted free access to Fable 5 from the launch date until June 22, after which the free access will be revoked.
Merely days after their release, the U.S. government imposed export controls on Fable 5, citing national security concerns.
Anthropic was instructed to suspend access to Fable 5 and Mythos 5 for all "foreign nationals," including non-U.S. individuals and employees of companies abroad.
Facing challenges in instantly differentiating access by nationality, the company opted for a "global full-scale withdrawal" of both models.
The official statement cited the trigger as a reported narrow-domain jailbreak method that could bypass safety guardrails and allow the model to invoke underlying capabilities in cybersecurity scenarios, a claim Anthropic disputes but has complied with.
In the short term, this analysis suggests that受阻 access to Fable 5 for domestic users will accelerate the回流 of demand towards local models and private deployments.
From a medium- to long-term perspective, the investment shift from "renting intelligence" to "building sovereign capability" is expected to elevate demand for local computing power and models.
Potential risks include slower-than-expected technological development and intensifying market competition.