Anthropic shipped Claude Fable 5 on June 9 with a genuinely impressive spec sheet: 1 million token context window by default, 128K output tokens per response, always-on adaptive thinking, and free access for Pro, Max, and Team subscribers through June 22. Three days later, US Commerce Department export controls disabled Fable 5 — along with Mythos 5 — for foreign nationals. Same week. Same model.
If you’re building anything on top of Anthropic’s API, this is worth understanding clearly.
What Fable 5 Actually Does
Fable 5 is Anthropic’s first publicly available Mythos-class model — the tier above Opus. The headline numbers are real and relevant:
1 million token context window. That’s roughly 750,000 words in a single prompt. You can feed it an entire codebase, a full year of customer support transcripts, or a research archive and ask a single coherent question across all of it. Previous Opus models topped out at 200K tokens.
128K output tokens. About 90,000 words in a single response. This matters for anyone generating long-form content, detailed reports, or complex code — tasks where previous models would hit the ceiling partway through.
Always-on adaptive thinking. Earlier models toggled extended thinking as an optional mode. Fable 5 applies it by default, which Anthropic says significantly improves performance on multi-step reasoning tasks.
On the API, Fable 5 runs at $10 per million input tokens and $50 per million output tokens. There’s a catch: Fable 5 and Mythos 5 use a new tokenizer that counts roughly 30% more tokens than models before version 4.7. If you’re cost-benchmarking against older Claude models, that difference adds up.
Stripe tested Fable 5 internally and reported it “compressed months of engineering into days” on a specific project. The model also handles vision tasks well enough to rebuild web apps from screenshots.
The Export Control Situation
On June 12, the US Department of Commerce issued export controls that suspended access to Fable 5 and Mythos 5 for foreign nationals. This includes users outside the US as well as foreign nationals working inside the US.
This isn’t Anthropic-specific — it reflects a broader US government policy around frontier AI models and national security. But the timing was brutal: three days after launch, a major portion of the API’s international user base lost access to the newest and most capable model in the lineup.
What does “suspended” mean in practice? Users in affected categories see Fable 5 unavailable in the model selector. The API returns an access error if claude-fable-5 is specified directly. Requests don’t silently fall back to Opus 4.8 by default — teams need to handle that fallback explicitly in their code.
Anthropic’s platform does include an optional fallbacks parameter for the API, which lets you specify alternate models if a request is refused or blocked. If you’re building internationally-facing tools on the Anthropic stack, that parameter is now load-bearing.
The Bigger Business Continuity Issue
Here’s the thing that should concern anyone running an AI-dependent stack: a model you’re using can become inaccessible — not because of a technical failure, but because of a policy decision made at the government level, announced three days after launch.
This is different from a typical software deprecation. Deprecations come with months of warning. Export controls don’t.
The practical lesson isn’t “don’t use Anthropic.” Fable 5 is genuinely the best model available on most benchmarks right now, and for users who can access it, the 1M context window opens up workflows that weren’t viable before. The lesson is that your AI stack should treat any single frontier model as potentially unavailable at any moment — and build accordingly.
That means: fallback models specified explicitly, not assumed. Context-heavy workflows redesigned so they can run on a smaller context window if needed. And a clear internal answer to the question “what do we do if our primary model goes offline tonight?”
For non-US teams evaluating their AI infrastructure: Anthropic’s export situation is worth factoring into vendor selection alongside cost and capability. It won’t be the last time this comes up.
For teams that retained access — including US-based subscribers on Pro, Max, and Team plans — Fable 5 is available and worthwhile. The free access window runs through June 22, then shifts to usage credits.
Want to go deeper on building with AI agents? Check out the AI Agent Lab → https://bit.ly/aiagentslab
