Claude Fable 5 access is restored and the model is generally available. Anthropic says the June export controls were lifted on June 30, 2026 and Fable 5 returned globally on July 1. Current documentation lists Fable 5 on the Claude API and supported cloud platforms at $10 input / $50 output per million tokens, with a 1M-token context window and up to 128K output. Claude Mythos 5 is also available, but only to approved Project Glasswing customers. The interruption described by this URL is a closed historical incident, not the current service status.
claude-fable-5 for generally available Fable access; Mythos 5 remains a separate limited-release model for approved Project Glasswing organizations.Claude Fable 5 and Mythos 5 status now
Anthropic’s current models overview calls Claude Fable 5 its most capable widely released model, intended for demanding reasoning and long-horizon agentic work. It is generally available through the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry. The API model ID is claude-fable-5.
Claude Mythos 5 shares Fable 5’s underlying capabilities, specifications and base pricing, but it does not include Fable’s safety classifiers. Mythos access is limited to approved Project Glasswing customers; organizations without that access can use the generally available Fable 5 model.
| Current fact | Claude Fable 5 | Claude Mythos 5 |
|---|---|---|
| Availability | Generally available through the Claude API and supported cloud platforms | Limited release to approved Project Glasswing customers |
| API model ID | claude-fable-5 | claude-mythos-5 |
| Context / max output | 1M / 128K tokens | 1M / 128K tokens |
| Base API price | $10 input / $50 output per 1M tokens | $10 input / $50 output per 1M tokens |
| Safety classifiers | Yes; can decline specified request categories | No Fable safety-classifier layer |
| Data retention | 30 days; no zero data retention | 30 days; no zero data retention |
For the evergreen background behind the name and the restricted program, see what Claude Mythos 5 is. For lineup and selection rather than incident status, use the current Claude model comparison.
The closed timeline: June 9 to July 1, 2026
The service interruption lasted less than three weeks. This chronology is historical; every event is dated so it cannot be mistaken for today’s availability.
| Date | What happened | Status after the event |
|---|---|---|
| June 9, 2026 | Anthropic released Fable 5 and Mythos 5. Fable was the safeguarded general-use product; Mythos went to a small group of defensive-cybersecurity partners. | Both models launched |
| June 12, 2026 | The US government applied export controls. Because nationality could not be verified reliably in real time, Anthropic paused both models for all users. | Access temporarily unavailable |
| June 30, 2026 | Anthropic announced that the export controls had been lifted after it developed and tested an improved safety classifier. | Redeployment announced |
| July 1, 2026 | Fable 5 returned globally. Mythos 5 access was restored within its approved Project Glasswing channel. | Access restored |
Why access was paused, and what changed
Anthropic’s official account says the June 12 controls followed a report about bypassing part of Fable 5’s cybersecurity safeguards. The company paused access because the government restriction took effect immediately and it lacked a reliable real-time way to verify nationality.
During the interruption, Anthropic reviewed the report with government and industry partners and trained an improved safety classifier. Anthropic says the new classifier blocks the reported technique in more than 99% of cases, while acknowledging that the stricter margin can also flag legitimate coding and debugging requests. The company then announced on June 30 that the controls had been lifted and redeployed Fable 5 on July 1.
The incident explanation and safeguard-performance figures above are Anthropic’s account. They are included to explain the closed event, not as an independent audit of the government decision or classifier.
Fable 5 API, pricing and model limits now
Fable 5 is a current API model rather than an archived launch. Anthropic’s official model overview and pricing pages list the following production facts:
- Model ID:
claude-fable-5. - Context window: 1 million tokens by default.
- Maximum output: up to 128,000 tokens per request.
- Base price: $10 per million input tokens and $50 per million output tokens.
- Thinking: adaptive thinking is always on and cannot be disabled; use the
effortparameter to guide depth. - Thinking display: raw chain of thought is never returned. The default display is omitted; request
"summarized"for a readable summary.
Those list prices do not include prompt-caching rates, Batch API discounts or provider-specific charges. For the subscription and credits angle, see our separate Fable 5 pricing and usage-credits guide.
Minimal current API call
from anthropic import Anthropic
client = Anthropic()
response = client.messages.create(
model="claude-fable-5",
max_tokens=2048,
messages=[{"role": "user", "content": "Review this migration plan."}],
)
if response.stop_reason == "refusal":
print("Request declined by a Fable 5 safety classifier")
else:
print(response.content[0].text)
Refusals are HTTP 200 responses, not API errors
Fable 5 includes classifiers for categories such as cyber, bio, frontier-model development and attempts to extract internal reasoning. When a classifier declines a request, the Messages API returns a successful HTTP 200 response with stop_reason: "refusal". Monitoring that only watches 4xx and 5xx errors will miss these events.
The response can include a stop_details object describing the category and a human-readable explanation. Applications should branch on stop_reason, not parse explanation text. If a refusal arrives before any output, Anthropic says the request is not billed and does not count against rate limits; a mid-stream refusal bills the input and output already produced.
Three fallback paths
- Server-side fallback: the Claude API and Claude Platform on AWS can retry a refusal automatically with the beta
fallbacksparameter. - SDK middleware: Anthropic’s refusal-fallback middleware works client-side across platforms.
- Manual retry: custom integrations can retry on another model and use Anthropic’s fallback-credit mechanism to avoid paying prompt-cache costs twice.
Anthropic’s current examples use Claude Opus 4.8 as a fallback target. A fallback is application behavior for a declined request; it does not mean Fable 5 itself is unavailable.
Data retention and zero data retention
Fable 5 and Mythos 5 carry a 30-day data-retention requirement and are not available under zero data retention. This is a model-specific term and should be reviewed before sending regulated, confidential or contractually restricted data.
The retention constraint is independent of the June incident. Access is restored, but organizations that require ZDR still need a different model or an approved data-handling design.
What deployment teams should do now
- Update availability checks. Remove temporary logic that assumes
claude-fable-5is globally offline. - Handle refusals explicitly. Emit metrics for
stop_reason: "refusal"and for responses served by fallback models. - Choose a fallback deliberately. Test the same prompt set on Fable 5 and the fallback model, including tool use and multi-turn history.
- Budget thinking and output together. Thinking tokens count toward
max_tokensand output billing even when the visible thinking display is omitted. - Review retention. Confirm the 30-day requirement fits your data policy before production use.
- Keep model selection configurable. Availability incidents are a reason to isolate model IDs and routing policy from application code.
Current status, pricing, API behavior and retention were verified on July 18, 2026 against the official Anthropic sources below. The June interruption is retained only as a closed, dated historical record.
Frequently asked questions
Is Claude Fable 5 available right now?
Yes. Claude Fable 5 access was restored on July 1, 2026, and Anthropic’s current documentation lists it as generally available through the Claude API and supported cloud platforms.
Is Claude Mythos 5 generally available?
No. Claude Mythos 5 is available only in limited release to approved Project Glasswing customers. Claude Fable 5 is the generally available model with the same underlying capabilities, specifications and pricing.
Why was Fable 5 access paused in June 2026?
The US government applied export controls on June 12 after a report about bypassing part of Fable 5’s cyber safeguards. Anthropic paused access for all users because it could not verify nationality reliably in real time.
When was Claude Fable 5 access restored?
Anthropic announced that the controls were lifted on June 30, 2026 and restored global Fable 5 access on July 1, 2026.
How do Claude Fable 5 and Mythos 5 differ?
They share capabilities, a 1M-token context window, 128K maximum output and $10/$50 pricing. Fable 5 includes safety classifiers and is generally available; Mythos 5 does not include those classifiers and is limited to approved Project Glasswing customers.
How does a Fable 5 API refusal work?
A classifier refusal is a successful HTTP 200 response with stop_reason set to refusal. Applications should detect that value and can retry through server-side fallback, SDK middleware or a manual fallback flow.
Does Claude Fable 5 support zero data retention?
No. Anthropic lists Fable 5 and Mythos 5 as covered models with a 30-day data-retention requirement, so neither is available under zero data retention.
Official sources
- Anthropic — Redeploying Fable 5. Official interruption, safeguard and restoration timeline.
- Anthropic — Introducing Claude Fable 5 and Claude Mythos 5. Current availability, model IDs, specifications, refusals and retention.
- Anthropic — Models overview. Current lineup, platform availability, pricing and limits.
- Anthropic — Claude API pricing. Base token and prompt-caching rates.
- Anthropic — Refusals and fallback. HTTP response shape, fallback methods and billing behavior.
- Anthropic — Fallback credit. Manual fallback billing and cache-cost treatment.
- Anthropic — API and data retention. Model-specific retention rules.
- Claude Help Center — Covered Models. Fable 5 and Mythos 5 retention status.
- Anthropic — Adaptive thinking. Fable thinking mode, display and token behavior.
- Anthropic — Effort. Controlling reasoning depth on adaptive-thinking models.
- Anthropic — Prompting Claude Fable 5. Current prompting guidance for Fable workloads.
