Claude Fable 5.1 is Anthropic’s general-release Mythos-tier model, available through Claude Pro, Max, Team, and Enterprise plans and Claude Code — after a nine-day worldwide shutdown in June 2026 triggered by a US export-control order over a disputed jailbreak. The September 1 update pushed Claude Code to the top of WebDev Arena’s front-end rankings. Pricing has held at $10/$50 per million tokens since June, unusually stable for a model this new.
Most model reviews on this site cover a launch and a benchmark. Fable’s story includes a US government shutdown order, a public dispute over what actually happened, and a relaunch with new safety machinery — all inside about three weeks. That history is why this review leads with the timeline instead of the specs sheet.
Nine Days That Made Fable Different From Every Other Anthropic Model
| Date | Event |
|---|---|
| Jun 9, 2026 | Fable 5 launches worldwide as Claude Code’s new default model for Pro/Max plans. Mythos 5 launches the same day, restricted to Project Glasswing cybersecurity partners only. |
| Jun 12, 2026 | US Commerce Department issues an export-control directive suspending access to both models for every foreign national, anywhere, including foreign-national Anthropic employees — citing an alleged jailbreak. |
| Jun 26, 2026 | Mythos access partially restored for some vetted US organizations. |
| Jun 30, 2026 | Commerce lifts the export controls entirely. |
| Jul 1, 2026 | Fable 5 restored worldwide with a new safety classifier in place. |
| Jul 20, 2026 | Fable 5 becomes a standard plan feature rather than a time-limited promotion. |
| Sep 1, 2026 | Fable 5.1 and Mythos 5.1 released. |
For three weeks, Fable simply wasn’t usable for anyone outside the US, and even US-based foreign nationals working at Anthropic itself were cut off. That’s a different kind of outage than a rate-limit change or a pricing update — it’s the first time an Anthropic model has intersected export-control law this directly.
Two Accounts of What Happened, and Neither Side Fully Settled It
The government’s directive didn’t spell out its specific concern in writing. Anthropic’s own account says its understanding was that officials had identified a way to bypass Fable’s guardrails — a technique the company says it reviewed and found reproduced a small number of already-known, minor vulnerabilities that other publicly available models could also surface, not a novel capability unique to Mythos. White House adviser David Sacks publicly said Anthropic had refused to fix the issue; Anthropic disputed that characterization and said it hadn’t received disclosure of any jailbreak that produced a harmful result.
Both positions are on the record and neither has been independently adjudicated in public. What is confirmed: the Commerce Department’s own AI safety institute (CAISI) reviewed Anthropic’s post-restoration fix and agreed the safeguards were strong, and the controls were lifted 18 days after they were imposed.
Fable vs. Mythos: The Split You’ll Actually Deal With
The naming causes real confusion, similar to the Nano Banana naming trap we covered in our Google AI models breakdown — a newer-sounding name doesn’t always mean more access.
- Fable is the version you can actually get. It shares Mythos’s underlying model but ships with safety classifiers that can decline high-risk requests (cybersecurity, biological, and LLM-R&D categories specifically) and reroute flagged prompts to Claude Opus 4.8 with a notice.
- Mythos is the same base model without those extra guardrails, and it’s only available to vetted organizations through Anthropic’s Project Glasswing program — mostly cybersecurity and life-sciences partners. If you’re reading this as a freelancer or small-business owner, you will not be offered Mythos, and that’s by design, not a paywall.
Practically: every mention of “Claude Fable” in a pricing page, a Claude Code changelog, or a benchmark chart refers to the model you can subscribe to. Mythos exists mostly as context for why Fable has the guardrails it does.
What September 1 Actually Changed
Fable 5.1’s biggest headline is a WebDev Arena result: it took the #1 spot on both the overall and front-end leaderboards running inside Claude Code. Independent trackers don’t agree on the exact number — one September snapshot puts it at 1,762 Elo (a 73-point gain over Fable 5), another puts it at 1,765 with Fable 5 itself sitting lower at 1,628. The precise figure moves depending on which tracker’s update cycle you catch it on, but every independent source agrees on the ranking order.
What that same data doesn’t advertise as loudly: Fable 5.1 drops out of the top 10 entirely on fullstack tasks, where Alibaba’s Qwen3.8 Max takes the lead instead. “#1 overall” is doing a lot of quiet work here — it’s a front-end result, not a universal one, the same composite-score pattern our AI model rankings piece found with Kimi K3 hiding behind another model’s headline score.
Two changes matter more for everyday use than the leaderboard position:
- Data retention: Fable 5’s initial rollout carried a mandatory 30-day data retention requirement. Reporting on the 5.1 update describes that requirement as removed — a real point in Fable’s favor if retention policy affects your compliance obligations.
- Classifier friction: the post-shutdown safety classifier blocks the reported jailbreak technique in more than 99% of attempts by Anthropic’s own account, but coverage of the July relaunch also noted it produces more false-positive interruptions on ordinary coding tasks than before. That trade-off doesn’t appear to have been reversed with 5.1 — expect the occasional routine request to get rerouted to Opus 4.8 with a notice, not a silent failure.
Pricing Hasn’t Moved Since Launch
Compared to the pricing swings we’ve tracked elsewhere on this site — Gemini’s promotional Flash pricing set to double on January 1, 2027, or DeepSeek’s “permanent” cut that got restructured into surge pricing within months — Fable’s rate card is remarkably boring:
| Model | Input $/M tokens | Output $/M tokens | Context window |
|---|---|---|---|
| Fable 5 (launched Jun 9, 2026) | $10 | $50 | 1M tokens |
| Fable 5.1 (launched Sep 1, 2026) | $10 | $50 | 1M tokens |
Three months, a government shutdown, and a capability jump later, and the API sticker price hasn’t changed. That’s genuinely unusual for a model in this news cycle — most of what we’ve covered this year moves in the other direction.
Who Should Actually Use This
If you’re calling the API directly, $10/$50 per million tokens is roughly double Claude Opus 4.8’s blended rate — reserve it for long-horizon coding-agent work or genuinely hard reasoning, not routine drafting, the same “match the tier to the job” logic in our coding-agent harness piece. Most small-business and freelance users won’t touch the raw API at all: if you’re on a Claude Pro, Max, or Team plan, or running Claude Code against one of those plans, Fable 5.1 is likely already your default model for coding sessions with no setup required — check you’re running Claude Code v2.1.255 or later, since older client versions don’t reliably surface the model selector. For general writing and research work outside of coding, our existing Claude AI review and Claude vs. ChatGPT writing comparison still cover the day-to-day assistant experience more directly than this review does.
Frequently Asked Questions
Is Claude Fable 5.1 safe to use after the shutdown?
The US Commerce Department’s own AI safety institute reviewed Anthropic’s post-restoration safeguards and agreed they were strong before the export controls were lifted. The underlying jailbreak dispute itself was never fully resolved in public, and Anthropic and the government gave different accounts of its severity.
What’s the difference between Claude Fable and Claude Mythos?
They share the same base model. Fable adds safety classifiers that can decline or reroute high-risk requests and is available on standard Claude subscriptions. Mythos lacks those extra guardrails and is restricted to vetted organizations through Anthropic’s Project Glasswing program — most users will never be offered it.
Why was Claude Fable banned for three weeks?
The US Commerce Department issued an export-control directive on June 12, 2026, citing national security concerns tied to a reported jailbreak technique. Anthropic disputed how serious the technique actually was. The controls were lifted June 30, and access returned worldwide July 1.
How much does Claude Fable 5.1 cost?
$10 per million input tokens and $50 per million output tokens on the API — unchanged from Fable 5’s original June 2026 price. Most subscribers access it through a Claude Pro, Max, Team, or Enterprise plan rather than paying per token directly.
Is Fable 5.1 better than Fable 5?
On WebDev Arena’s front-end and overall rankings, yes — it holds the #1 spot as of September 2026. On fullstack coding tasks specifically, it drops out of the top 10 entirely, with Alibaba’s Qwen3.8 Max leading instead. It also removed Fable 5’s mandatory 30-day data retention requirement.
Can I get access to Claude Mythos instead of Fable?
Only if your organization is accepted into Anthropic’s Project Glasswing program, which is aimed at vetted cybersecurity and life-sciences partners. There’s no consumer or standard-business path to Mythos access.
Shurah is the founder of AI Tools Daily, tracking pricing, licensing and policy changes across AI tools so readers can make decisions without wading through marketing claims themselves.
