Anthropic released a new model called Claude Fable 5 on June 9, calling it a “Mythos-class” model — the company’s most capable tier yet. But just three days after launch, the U.S. government imposed export controls that forced Anthropic to cut off access. The restrictions were lifted on June 30, and the model returned to users on July 1. Here is what this means for developers and startups in Uzbekistan, and how to actually use it.
What Fable 5 is, and what makes it different
According to Anthropic, Fable 5 and Mythos 5 are built on the same underlying model — the difference lies in their safety restrictions. Mythos 5 runs without additional filters and is available only to a select group of partners under Project Glasswing, mostly organizations working in cyber defense and critical infrastructure. Fable 5 has the same capabilities but comes with extra safety classifiers layered on top, which is why it was released for general use.
Anthropic states that Fable 5 is the most capable model it has ever made publicly available: it delivers state-of-the-art performance in software engineering, knowledge work, vision, scientific research, and more. The longer and more complex a task, the greater Fable 5’s advantage over other models — a trait that matters most for multi-stage, autonomous projects, such as coding tasks that run for several days.
Where and how to use it
Fable 5 is accessible through the following channels:
- Claude.ai — for Pro, Max, Team, and select Enterprise plans.
- Claude API and Claude Platform — model string
claude-fable-5. - Cloud platforms — also available via Amazon Bedrock, Google Cloud, and Microsoft Foundry.
- Claude Code and Claude Cowork — for agentic workflows.
Pricing follows the standard rate: $10 per million input tokens and $50 per million output tokens (1.1x for workloads requiring U.S.-only infrastructure). The context window is 1 million tokens by default, with output capped at up to 128,000 tokens per request.
Through July 7, eligible Claude subscribers (Pro, Max, Team) can use up to 50% of their weekly usage limit on Fable 5 at no extra cost; after that, usage shifts to a credit-based system.
The trouble started in cybersecurity. Amazon researchers found a way to bypass Fable 5’s safeguards, prompting it to identify software vulnerabilities and, in one case, to produce code demonstrating how a vulnerability could be exploited. This led the U.S. Department of Commerce to issue an order on June 12 restricting access for non-U.S. nationals. Since Anthropic had no way to verify user nationality in real time, it suspended the model entirely, for all users.
Anthropic pushed back on the government’s move, noting that the same vulnerabilities could be identified by less capable models, including Opus 4.8, GPT-5.5, and Kimi K2.7 — meaning Fable 5 offered no unique offensive capability. On June 30, Commerce lifted the restrictions, and Anthropic rolled out updated cybersecurity safeguards, including a classifier the company says blocks the relevant jailbreak technique in more than 99% of attempts. It also launched a HackerOne program for security researchers to report new jailbreaks.
Strengths and limitations
Strengths: autonomous execution of long, complex tasks; the ability to write its own tests and check its own code; understanding of tables and diagrams embedded in documents; handling multi-stage research and analytical work with minimal oversight.
Limitations: a significant share of requests related to biology, chemistry, and cybersecurity are automatically rerouted to the less capable Opus 4.8 model — users aren’t charged Fable pricing for these, but the output also isn’t Fable-level. In addition, Fable 5 and Mythos 5 are subject to a 30-day data retention policy and are not available under zero data retention — a relevant consideration for companies handling sensitive corporate data.
The return of Fable 5 signals a broader shift toward handing long-running, complex development work to AI agents — a trend that’s directly relevant to Uzbekistan’s outsourcing and product companies. At the same time, the episode is a reminder of how quickly AI models can get pulled into geopolitical oversight: a three-week suspension could have meant an unexpected disruption for teams working on international projects. It’s also a caution against building critical workflows around a single model.
