Release analysis
Claude Fable 5 and Mythos 5: what launched, what broke, and what it costs
Claude Fable 5 and Claude Mythos 5 are two access policies over a single model: Anthropic's most capable widely released system, launched on 9 June 2026 as the first tier in a new "Mythos-class" band that Anthropic positions above its Opus class. Fable 5 is the generally available version, shipped with safety classifiers; Mythos 5 is described by Anthropic as the same underlying model with those classifiers removed, restricted to approved organisations through a programme called Project Glasswing. What makes the launch worth a dated reading is not the capability claims but what happened next: within three days the pair was pulled offline entirely under US export controls, and it did not return to global availability until 1 July.
Dated analysis. This is a reading of the launch and its aftermath as the record stood on 18 July 2026. Product facts, pricing and the export-control timeline are drawn from Anthropic's own posts and mainstream press; the post-relaunch user complaints are reported claims, not results this site has reproduced, and are framed as such below.
What is "Mythos-class", and how do the two variants differ?
"Mythos-class" is Anthropic's name for a new capability tier it places above the existing Opus class, aimed at demanding reasoning and long-horizon agentic work. The important point for buyers is that Fable 5 and Mythos 5 are not two different models. Anthropic's platform documentation (checked 18 July 2026) states that Mythos 5 shares Fable 5's capabilities and differs only in that it ships without Fable 5's safety classifiers.
Those classifiers cover three dual-use domains: cybersecurity,
biology and chemistry, and suspected model-distillation attempts. When one
fires, the query is answered by Claude Opus 4.8 instead, and the API returns
stop_reason: "refusal" as an HTTP 200 response rather than an
error, naming the classifier that triggered. Anthropic states this happens in
fewer than
5% of Fable 5 sessions, so on its own figures more than nineteen sessions
in twenty reach the unrestricted model. Mythos 5 carries none of this and is
available only to Project Glasswing partners - initially cyberdefence and
select biology-research organisations - with access brokered through
Anthropic, AWS or Google Cloud account teams.
What does Fable 5 cost, and what are its limits?
Both variants are priced identically, and both carry the same context and output limits. Anthropic frames the price as less than half that of the earlier Claude Mythos Preview.
| Attribute | Fable 5 / Mythos 5 |
|---|---|
| Input price | $10 / M tokens |
| Output price | $50 / M tokens |
| Context window | 1,000,000 tokens |
| Max output | 128,000 tokens |
| Data retention | 30-day mandatory; no zero-data-retention option |
The retention line is not a footnote. Both models carry a mandatory 30-day retention window and are not offered under zero-data-retention terms, which the launch documentation confirms and which reporting flagged as a practical lockout for European companies operating under GDPR data-minimisation duties. If that constraint applies to you, the headline benchmark is beside the point until the retention terms change.
Why did Fable 5 go offline three days after launch?
On 12 June 2026, three days after launch, the US government applied export controls to both models. The trigger, per Anthropic's own redeployment notice, was that Amazon researchers had found a technique to bypass Fable 5's safeguards, enabling it to identify software vulnerabilities and, in one case, produce exploit-demonstration code. Because Anthropic could not reliably verify user nationality in real time, it suspended access for all users rather than only foreign nationals. The controls were lifted on 30 June, with Mythos 5 access restored to a set of approved US organisations on 26 June, and Fable 5 returned to global availability on 1 July. The model that headlined as a frontier release spent roughly its first three weeks either offline or restricted.
What went wrong after the relaunch?
Reception after 1 July was hostile, and it is worth separating the corroborated shape of the complaints from their unverified detail. Users reported Fable 5 consuming tokens at roughly twice the rate of prior models; one widely cited example described a $100 allowance exhausted in under nine minutes. A separate wave of "nerfing" complaints followed the re-release, with third-party testers reporting steep score drops. Anthropic and outside analysis attributed those drops not to a change in the underlying model but to more aggressive fallback to Opus 4.8 when the safety classifiers fired - the router, not the model, in Decrypt's framing. The precise benchmark figures circulating in that episode come from a single named test group and have not been verified against a primary publication, so treat the direction as reported and the numbers as unconfirmed.
The through-line is that Fable 5's real-world behaviour is governed as much by its routing and safeguard layer as by the base weights. A capability number measured with the classifiers quiet tells you little about a session where they are firing. If you are weighing the tier, the two questions that matter more than the launch table are whether the 30-day retention rules you out, and how often your own workload lands in the cyber, bio or distillation lanes that hand your query to Opus 4.8.
Questions people ask
What is the difference between Fable 5 and Mythos 5? They are the same underlying model under two access policies. Fable 5 is generally available and carries safety classifiers that redirect certain cyber, bio and distillation queries to Opus 4.8; Mythos 5 has those classifiers removed and is limited to Project Glasswing-approved organisations.
What does "Mythos-class" mean? It is Anthropic's label for a new capability tier positioned above its Opus class, aimed at demanding reasoning and long-horizon agentic work. Fable 5 is the first model in it.
How much does Fable 5 cost? $10 per million input tokens and $50 per million output tokens, with a one-million-token context window and up to 128,000 output tokens per request - the same figures as Mythos 5.
Why was Fable 5 restricted after launch? US export controls applied on 12 June 2026, after Amazon researchers bypassed Fable 5's safeguards. Anthropic suspended access for all users because it could not verify nationality in real time. The controls were lifted on 30 June and Fable 5 was globally available again on 1 July.
Can EU companies use it under GDPR? Both models carry a mandatory 30-day data-retention window with no zero-data-retention option, which reporting identified as a lockout for organisations bound by GDPR data-minimisation requirements. Confirm your own compliance position before committing.
Sources & caveats. Product, pricing and safeguard facts come from Anthropic's Fable 5 and Mythos 5 announcement (9 June 2026) and its platform documentation (checked 18 July 2026); the export-control timeline from Anthropic's redeployment notice, CNBC (30 June 2026) and Al Jazeera (1 July 2026). Post-relaunch user complaints are reported via Decrypt; the specific "nerfing" score drops are attributed to a single named test group and are not independently confirmed here. Vendor benchmark claims are excluded from this piece pending independent replication.