TL;DR: Anthropic launched Claude Fable 5 on June 9, 2026 – the first publicly available Mythos-class model and the most capable AI Anthropic has released. It landed in GitHub Copilot across every major IDE and was available to Pro+, Max, Business, and Enterprise tiers. Three days later it was suspended globally after a US government export-control directive targeted Mythos-class capabilities for foreign nationals. If you are outside the US, it is currently unavailable with no workaround. Claude Opus 4.8 is your best option in the meantime. Pricing when it returns: $10 per million input tokens, $50 per million output tokens.
Updated June 19, 2026: Since I first published this post, Fable 5 has been suspended globally and the reasons matter more than the disruption itself. Read on for the full picture.
Something quietly significant landed on June 9, 2026. Anthropic’s Claude Fable 5 went generally available inside GitHub Copilot and if you have been following the Mythos story since it broke in March, you will know this was a bigger deal than the usual model-picker update.
Then, three days later, it was gone.
Let me break down what happened, why the data retention story has a new chapter, and what this means for anyone like me, building from Wales, who was already planning their stack around Mythos-class capability.
What Is Claude Fable 5?
Fable 5 is the first publicly available model in Anthropic’s Mythos class, a capability tier that sits above Opus. That’s worth pausing on. Anthropic’s Opus models have been the top of the stack for a while now. Mythos changes that hierarchy entirely.
The original Claude Mythos model was kept off the public market for good reason. It demonstrated an unusually high ability to identify and exploit software vulnerabilities at a level matching elite human researchers. Anthropic ran it under Project Glasswing, a restricted programme for a small set of trusted partners. Fable 5 is effectively the same underlying model, wrapped in a classifier safety layer that makes it suitable for general release. That is not a compromise. That is genuinely impressive engineering.
The model itself is designed for long-horizon, autonomous work and not chat-style completions. In Anthropic’s internal benchmarks, Fable 5 completed equivalent autonomous coding workflows with fewer tool calls and lower token consumption than previous Opus-tier models. That matters directly for cost and latency in production pipelines.
Where You Could Use It – And Where You Cannot
On June 9, Fable 5 launched across the full GitHub Copilot surface: VS Code (chat, ask, edit, and agent modes), Visual Studio, JetBrains, Xcode, Eclipse, Copilot CLI, the GitHub Copilot cloud agent, the GitHub Mobile app on iOS and Android, and directly on github.com. Available to Copilot Pro+, Max, Business, and Enterprise tiers.
On June 12, three days later, GitHub added an editor’s note to the changelog: access to Claude Fable 5 had been suspended across all GitHub Copilot experiences, effective immediately.
The reason? Anthropic says it received a US government directive at 5:21pm ET on June 12, citing national-security export-control authority, ordering the suspension of access to Fable 5 and Mythos 5 for any foreign national, inside or outside the United States, including foreign-national Anthropic employees. Unable to enforce that restriction selectively in real time, Anthropic took the only practical option available and disabled both models for all customers globally.
All other Claude models in GitHub Copilot, including Opus 4.8, Sonnet 4.6, and Haiku 4.5, remain available and unaffected.
The Data Retention Picture – Now With Added Complexity
The original data retention story was already worth understanding carefully. Fable 5 departs from the Zero Data Retention model that other Claude models in GitHub Copilot operate under. Specifically: Anthropic retains prompts and outputs for up to 30 days to run the safety classifiers that make a Mythos-class model safe to release publicly. After 30 days, that data is deleted and it is not used to train Anthropic’s models.
That is a load-bearing architectural decision, not a privacy slip. The classifier layer that keeps Fable 5 from becoming a vulnerability-hunting tool for bad actors requires operational data to function. You cannot have the safety without the retention.
The suspension adds a second dimension to this. Flagged items can apparently be held for significantly longer under “safety investigations” and “legal purposes” exceptions. That is the clause Microsoft’s own legal team reportedly flagged when they quietly blocked Fable 5 from their internal GitHub Copilot instance, even while selling it to paying customers externally. The outer retention bound in those edge cases is not as clearly defined as the standard 30-day window and that ambiguity is exactly the kind of thing enterprise legal teams will anchor on.
Every other Claude model in Copilot, including Opus 4.8, Sonnet 4.6, and Haiku 4.5, continues to operate under full ZDR. Those remain the right choice for any org with hard data residency requirements.
Pricing
Fable 5 is billed at provider list pricing under GitHub Copilot’s Usage Based Billing model. Anthropic’s published list rate sits at $10 per million input tokens and $50 per million output tokens. Given the benchmark data on efficiency, with fewer tool calls and lower token consumption relative to Opus, the effective cost per task may well be lower than it looks at the headline rate. Pricing stands regardless of the current suspension and when access is reinstated, the rate structure is unchanged.
What This Means If You Are Outside the US
If you are building from the UK, or anywhere outside the US, the honest answer right now is that Fable 5 is not available to you and there is no workaround. Anthropic’s eligibility check is account-based, not IP-based, so a VPN will not help. The best non-Mythos option remains Claude Opus 4.8 (released May 28, 2026), available in all regions where Anthropic operates and still the strongest model in the Opus tier.
This is frustrating. It is also, given what Mythos-class models are capable of, not entirely surprising that frontier-AI export controls entered the picture. The speed at which this moved, from general availability to global suspension in 72 hours, is the part that should shape how you think about building on any single frontier model.
What This Means for My MVP
I have been building toward an application that leans heavily on autonomous, multi-step reasoning, which is exactly the profile Fable 5 is designed for. The Opus tier has been serviceable, but the ceiling was visible. Fable 5 removing that ceiling, even briefly, confirmed the direction is right. The suspension does not change the destination. It just means I am building on Opus 4.8 until the export-control picture stabilises.
The broader lesson here, and one I am baking directly into the MVP architecture, is that model availability is now an operational dependency in exactly the same way that API uptime is. You need fallback logic. You need to abstract your model selection layer. And you need to be honest with stakeholders that the frontier moves fast in both directions.
I will be posting more as this develops, both the regulatory story and the build. If you are navigating the same decisions outside the US, I would genuinely love to compare notes. Drop a comment below or find me on LinkedIn.
Further reading: GitHub Changelog – Claude Fable 5 (includes June 12 suspension note) – Anthropic commercial terms and data retention policy
