Last updated: June 29, 2026
Key Takeaways
- Real movement, but not for you yet. On June 26 the government cleared the restricted Mythos 5 model for a vetted set of US critical-infrastructure organizations. Claude Fable 5, the version subscribers and developers actually used, is still suspended worldwide on day 17.
- A broad return looks likely and possibly soon. A late-June report says officials expect Fable back within days, and traders price eventual restoration as near-certain. But no date is committed, near-term odds have cooled, and several "it's back" reports have already been retracted.
- Expect changed terms on return: metered usage credits rather than unlimited plan access, and a new ID-verification option (effective July 8) that may gate access. The capability you run on your own hardware is the version nobody can revoke.
On June 12, 2026, a model hundreds of millions of people had been using three days earlier vanished by government order. A US Commerce Department export-control directive forced Anthropic to disable Claude Fable 5 and its restricted sibling Mythos 5 for every customer worldwide within hours. Seventeen days later the picture has finally moved, though not the way most subscribers hoped: on June 26 the government let Mythos 5 back for a narrow list of vetted organizations, while Fable 5 stays dark for everyone.
The question filling search boxes has not changed: when does Fable 5 come back, and what does it look like when it does? No one outside Anthropic and the government knows the date, which has not stopped a flood of confident predictions, retracted "restoration" reports, and outright hoaxes. This piece separates confirmed fact from speculation on the timing, cost, and ID question, and labels guesses as such.
What changed: Mythos 5 is partially back
The first concrete government movement since the shutdown came on June 26, when Commerce Secretary Howard Lutnick sent Anthropic chief compute officer Tom Brown a letter narrowing the order. As reported by CNBC and Semafor, Lutnick wrote that appropriate safeguards were in place to let certain trusted partners use Mythos 5. Anthropic confirmed it the next day in an official post, saying the government had cleared its strongest cybersecurity model for a set of US organizations that defend critical infrastructure, and that it was restoring their access quickly while working toward general availability for Fable 5.
The scope is deliberately small. Reporting puts the cleared list near 100 organizations, many of them existing partners in Anthropic's restricted Project Glasswing program, plus their foreign-national staff and US government agencies and labs. Mythos 5 is the unrestricted, safeguard-free sibling of Fable 5; we covered how Glasswing gates it in our silent-safeguards breakdown. What the letter does not do is bring Fable 5 back: it went unmentioned, the original penalties stand, and ordinary subscribers, API developers, Claude Code users, and international customers stay on Opus 4.8, Sonnet 4.6, and Haiku 4.5. The partial restoration proves a negotiated path exists, but its narrowness is the message. Broad public access is the harder problem, not the easy one.
Is Fable 5 back yet?
No. Fable 5 has been suspended since June 12, new sessions default to Opus 4.8, and direct API calls to claude-fable-5 return errors. Independent trackers ping Anthropic's API on a one-minute loop and all read the same, that the model is not responding. Anthropic's own status page still lists the incident with no general-restoration entry.
Before you trust anything you read: one outlet published a detailed report that Fable 5 had returned in mid-June with nationality-based controls, then retracted it days later because it could not be substantiated. When fresh rumors claimed a specific Claude Code build had unlocked the model, Anthropic staff shot them down, one engineer saying the company was serving exactly 0 traffic to Fable 5. The model still appears in some mobile pickers and cloud catalogs, but selecting it fails or routes to Opus 4.8. That is a stale menu entry, not a soft launch. Treat any "Fable is back" claim that does not point to the status page or a working API call as noise.
Why it was pulled, in brief
You cannot reason about the return without the reason for the exit. The directive ordered Anthropic to block any foreign national, a category that under export-control law includes green-card holders, visa workers, and the company's own non-citizen employees. With no reliable way to verify citizenship in real time at the API layer, the only compliant move was to switch the models off for everyone. We covered the model, its pricing, and its data-retention terms in our Fable 5 and Mythos 5 breakdown.
The trigger is disputed, and the dispute drives the timeline. Anthropic's public statement describes a narrow technique that amounts to asking the model to read a codebase and fix software flaws, argues the same capability exists in other public models including GPT-5.5, and calls the order a misunderstanding. The government's side, first relayed by The Economist, sounded far more severe: the NSA director reportedly told a Senate briefing that Mythos breached nearly all of the agency's classified systems within hours. That account was later walked back, with the Associated Press and a follow-up Economist clarification establishing the exercise as a sanctioned test to find software flaws before attackers do, the model finding vulnerabilities fast but not shown to exploit them. The distinction sets the timeline: a patchable jailbreak gets fixed quickly, while a capability the model fundamentally has does not.
When will Claude Fable 5 come back?
The negotiation path
The most direct route is the dispute getting resolved, and after two static weeks there are signs of motion. The June 26 Mythos clearance is the first concrete step, and a late-June Axios report said insiders expect Fable 5 to follow within days, crediting Lutnick and Treasury Secretary Scott Bessent with defusing the standoff. Anthropic's managing director for international, Chris Ciauri, had said back on June 17 in Seoul that the models would return in the coming days. None of that is a committed date, and the status page has not moved for general users.
The structural markers that actually gate it
Beyond the back-channel talks, a few fixed dates are the real levers. The table below lays out the paths that could change Fable 5's status, with our read on each.
| Path or trigger | Date | What it would mean | Our read |
|---|---|---|---|
| Mythos 5 partial restoration | Done, June 26 | Mythos cleared for vetted US critical-infrastructure orgs | First concrete step; Fable not included |
| Direct talks with Commerce | Ongoing | Directive narrowed or withdrawn for Fable | Reportedly easing; plausible near-term |
| ID-verification policy live | July 8, 2026 | Could enable a US-verified-only restoration | Contested whether it resolves the order |
| Directive fully lifted for Fable | No date set | Clean restoration for everyone | Likely eventually; timing unknown |
Dates are fixed; outcomes are not. These are levers that could move Fable 5's status, not a schedule of guaranteed events. Verify the live position at status.claude.com before relying on any of it.
What the odds say
If you want a number, prediction markets are the least bad source, since traders price real money against the outcome. The picture is now split by timeframe. On Polymarket, where more than three million dollars has traded, eventual US restoration is priced as near-certain, with the August and December contracts in the mid-90s percent. The near-term cooled the other way: after the partial restoration showed how narrow the first step was, odds of a broad return before July 1 slipped into the mid-40s, down from the high-50s a week earlier. Kalshi has run higher near-term, and all of it moves daily.
Our honest bottom line: a return is very likely and the base case is weeks not months, but a specific date is genuinely unknowable. Anyone who gives you one is guessing.
What will it cost when it returns?
Here the speculation rests on firmer ground, because the pricing was public before the shutdown. On the API, Fable 5 launched at $10 per million input tokens and $50 per million output tokens, roughly double Opus 4.8 and the most expensive model Anthropic ships, below the restricted Mythos Preview tier at $25 and $125. On subscriptions it was free on Pro, Max, Team, and seat-based Enterprise plans from June 9 through June 22, but counted as about twice Opus usage against your limits. It was set to move to metered usage credits on June 23; the suspension overtook that cliff, and Anthropic has announced no billing adjustment for the outage.
So what does the return look like on your invoice? The strong signal is metered usage credits rather than unlimited plan access, possibly with a bundled weekly allotment that draws on credits once spent. One prediction market puts the odds of pure pay-per-token pricing, with no subscription access at all, at under a third, implying most traders expect some bundled access back. The longer-term direction points down, since frontier pricing consistently falls after launch, often 30 to 50 percent within six months. Read today's rate as a launch ceiling, not the resting price.
Will you have to verify your ID?
This is the most consequential change for ordinary users, and it is partly confirmed. Anthropic updated its policy so consumer accounts (Free, Pro, and Max) can be asked to verify identity in certain circumstances by uploading a government-issued ID and submitting facial-geometry data, effective July 8, 2026. Business and API customers are exempt, and verification runs through a third-party vendor, Persona.
The link to Fable is a theory, not a fact. The optimistic reading, laid out by CIO, is that ID checks let Anthropic open access to verified US persons without waiting for the government to lift the order. The skeptical reading is that an ID check does not address the export hurdle at all, since the directive is about nationality. Both are defensible; treat a US-verified-only return as a plausible path, not a settled plan. The privacy cost is worth stating plainly: Persona has been reported checking users' biometric data against government watchlists, a practice that previously led Discord to drop a partnership. A chatbot asking for your passport and a live selfie is a real shift in what using a frontier model requires.
What to use while you wait
The gap Fable left has filled with capable, heavily marketed alternatives, each carrying a caveat the marketing skips. Sakana's Fugu pitched itself shoulder-to-shoulder with Fable, but on its own numbers it generally trails Fable 5, and it is an orchestrator you cannot self-host or inspect; we ran the benchmarks in our Fugu versus Fable 5 comparison. OpenRouter's Fusion makes a similar "Fable-level at half the price" claim that holds only for one configuration, covered in our Fusion analysis. Moonshot's Kimi K2.7 ships open weights at a fraction of Fable's per-token cost, though the full model needs hundreds of gigabytes of memory.
A safety note: the block sits server-side, enforced by government directive, so no file, prompt, hidden parameter, or unofficial mirror reactivates Fable 5. Anything promising a back door is a hoax at best and a way to hand your prompts and credentials to a stranger at worst.
The bigger lesson: rented capability can be revoked
Strip away the specifics and this is the cleanest demonstration yet of the idea that runs through everything we cover: whoever controls the infrastructure controls the experience. A model hundreds of millions depended on was gone in an afternoon, by decree, with no user action and no say in the matter. The episode has not gone unchallenged: more than 100 cybersecurity leaders, among them Alex Stamos and Chris Wysopal, signed a letter urging officials to lift the restrictions, arguing the ban strips defenders of a useful tool without slowing attackers. That argument may prevail, but it is one being had over your head, about access you do not control.
The hedge that cannot be revoked, repriced, or ID-gated is an open-weight model on hardware you own. No open model matches Fable-class capability on the hardest long-horizon agentic work, but for the work most people run day to day, document search, summarizing, drafting, coding help, and home automation, open models cleared the sufficiency bar a while ago. The weights live on your disk, the terms cannot change underneath you, and nothing you type leaves your network. If this episode moved local AI from curiosity to plan, our mini PC guide for local AI covers the always-on hardware and our best local models by VRAM rundown covers the software. Owning your own floor stops today's "is Fable back" question from dictating whether you can work.
Frequently Asked Questions
Is Claude Fable 5 back yet?
No. As of June 29, 2026, seventeen days in, Fable 5 is still offline for general users. API calls error, sessions default to Opus 4.8, and the status page shows no general restoration. Anthropic staff confirmed in late June that the company is serving zero traffic to the model.
Is Claude Mythos 5 back?
Partly. On June 26 the government cleared Mythos 5, the restricted sibling model, for a vetted set of US critical-infrastructure organizations and certain government partners. That clearance does not cover Fable 5, ordinary subscribers, API developers, or any international users.
When will Fable 5 realistically return?
No date is confirmed. A late-June report said officials expect a broad return within days, and prediction markets price eventual restoration as near-certain while putting a return before July 1 in the mid-40s percent. Weeks, not months, is the realistic read, but timing hinges on a policy dispute no one outside the room can call.
Will I have to upload my ID to use Fable 5?
Possibly. From July 8, Anthropic can ask consumer accounts to verify identity with a government ID and facial-geometry data in certain cases; business and API customers are exempt. Whether it gates Fable specifically, or even resolves the export issue, is not confirmed.
Can I run Fable 5 locally, or is there an open version?
No. The weights are closed and inference runs only on Anthropic's servers, so there is no legitimate local version and no file or prompt unlocks the suspended model. The nearest path is running open-weight models through a tool like Ollama on capable hardware, which no one can revoke.
