Claude Fable 5 Pricing, Access and First Benchmarks

01Five key points

• Anthropic released Claude Fable 5 on June 9, 2026, the first "Mythos class" model open to everyone.
• API pricing is $10 per million input tokens and $50 per million output tokens, with a 1 million token context window and up to 128,000 output tokens.
• Sensitive requests (offensive cybersecurity, biology, chemistry) get answered by Claude Opus 4.8 instead. More than 95% of sessions are never affected.
• Anthropic puts it at the top of most published benchmarks, including 80.3% on SWE-bench Pro according to the specialist press.
• It is available through the Anthropic API, claude.ai, OpenRouter, Amazon Bedrock, Vertex AI and Microsoft Foundry.

Update of June 12, 2026. Three days after this launch, a US government directive forced Anthropic to suspend Fable 5 and Mythos 5 for every customer, worldwide. The suspension of Fable 5 explains what happened and which models to use instead. The launch details below describe the model as it was announced on June 9.

Update of October 16, 2026. The suspension ended after the export controls were lifted on June 30, 2026. Anthropic restored Fable 5 for users worldwide on July 1, then released Claude Fable 5.1 on September 1, 2026 at the same API price of $10 per million input tokens and $50 per million output tokens.

0202The Mythos class sits one step above Opus

Until now, Opus models were the top of the Claude lineup. Anthropic has added a higher tier called Mythos, which it has tested since April 2026 with a small group of partners under the name Claude Mythos Preview. Fable 5 is the public version of this class. It runs on the same base model as Claude Mythos 5, with stronger safeguards that let Anthropic release it at scale.

Mythos 5 itself stays limited to selected organizations. Anthropic is rolling it out first to cyber defenders and infrastructure operators through Project Glasswing, a program run with the US government. About 150 organizations in more than 15 countries are due to join it.

0303The published figures

Anthropic says the model is state of the art on almost every benchmark it tested. It adds that the gap widens on long tasks. The figures below come from the official announcement and from the specialist press, The Decoder in particular. We have not reproduced them ourselves yet, so treat them with the same caution as any vendor's numbers.

Scores reported by The Decoder on June 9, 2026 (higher is better).

Model	SWE-bench Pro	FrontierCode
Claude Fable 5	80.3%	29.3%
Claude Opus 4.8	69.2%	13.4%
GPT 5.5	58.6%	5.7%
Gemini 3.1 Pro	54.2%	not published

Beyond benchmarks, the most telling account comes from Stripe. The company reports that Fable 5 completed a migration of a 50 million line Ruby codebase in a single day. By Stripe's estimate, a human team would have needed more than two months for the same job.

Anthropic also highlights vision. The model can pull precise values out of scientific figures and rebuild the source code of a web app from plain screenshots. On endurance, it works on its own for longer than any earlier Claude model and keeps track of the task across millions of tokens.

0404Safeguards that hand off to Opus 4.8

What sets Fable 5 apart is its safety setup. Classifiers watch three areas of requests. One is offensive cybersecurity. Another is biology and chemistry. The last is distillation, meaning extracting the model's skills to train a competitor. When a classifier fires, Claude Opus 4.8 handles the answer instead of Fable 5. The user is told when this happens.

According to Anthropic, more than 95% of sessions never trigger a handoff. The company also says an external bug bounty program found no universal jailbreak in more than 1,000 hours of testing. In practice, the switch to Opus 4.8 stays the exception rather than the rule.

Businesses should know that Anthropic keeps traffic from Mythos class models for a mandatory 30 days. The company says this retention serves security purposes only. If you process sensitive data, check this point with your data protection officer before you adopt the model.

0505Pricing and access

On the API, Fable 5 costs $10 per million input tokens and $50 per million output tokens. Anthropic says that is less than half the price of Mythos Preview. The 90% discount on cached input tokens still applies. The official documentation lists a 1 million token context window and up to 128,000 output tokens, under the model ID claude-fable-5.

On claude.ai, the rollout happens in two stages. From June 9 to 22, 2026, the model is included at no extra cost in the Pro, Max, Team and Enterprise plans. After that it will require usage credits while Anthropic expands its compute capacity. The Decoder adds that each Fable 5 request counts double against subscription quotas. The model is also available today on OpenRouter, Amazon Bedrock, Vertex AI, Microsoft Foundry and in GitHub Copilot.

0606What it changes for web professionals

Three of the announced capabilities look directly useful if you build or optimize websites. The first two concern what the model can actually do on a project. The third concerns what each task will cost you once the bill arrives.

• Long, autonomous tasks. A full site audit, an internal linking overhaul or a migration touches hundreds of pages. The announced gains on this kind of project (the Stripe example first) target exactly this sort of deep work.
• Vision. Rebuilding an interface's code from screenshots opens up practical uses. You can analyze a competitor's page, turn a mockup into working front-end code or make sense of a Search Console chart.
• Output pricing. At $50 per million output tokens, Fable 5 is best kept for tasks where it makes a real difference. A meta description or a simple answer works fine with a lighter model. Picking the right model for each task is how OpenTahr is built.

Stripe said Fable 5 compressed months of engineering work into a few days.

As reported in Anthropic's official announcement of June 9, 2026.

0707Try it with your own key

You don't need a subscription to test Fable 5, because an API key is enough. There are two ways to get one. The first goes straight to Anthropic, while the second goes through OpenRouter and its large model catalog.

• Anthropic key. Create a key in the Anthropic console, then call the claude-fable-5 model. You pay for what you actually use, with no middleman.
• OpenRouter key. One key gives you Fable 5 (model ID anthropic/claude-fable-5) plus more than 340 other models. That makes it easy to compare answers to the same prompt.

If you use OpenTahr, your key stays stored in your browser and calls go straight to the provider. The key storage details are on the home page. One tip holds everywhere. Set a spending cap with your provider before you start long tasks, because Fable 5's output tokens add up fast.

0808Frequently asked questions

Is Claude Fable 5 free?

No. The API costs $10 per million input tokens and $50 per million output tokens. On claude.ai, the model is included in paid plans from June 9 to 22, 2026. After that date, using it there will require usage credits.

What is the difference between Fable 5 and Mythos 5?

They share the same base model. Fable 5 is the public version, with stronger safeguards. Mythos 5 lifts some of those safeguards and stays limited to selected partners, in cyber defense in particular through Project Glasswing.

Is it available in France?

Yes. Access from France works the same way as it does anywhere else. You can reach the model through the Anthropic API, claude.ai, OpenRouter, Amazon Bedrock, Vertex AI or Microsoft Foundry.

Does it replace Claude Opus 4.8?

No. Opus 4.8 stays in the catalog and costs less. It also serves as the fallback model for the sensitive requests sent to Fable 5. Anthropic positions Fable 5 for long, complex tasks.

0909References

1. Anthropic, "Claude Fable 5 and Claude Mythos 5", official announcement, June 9, 2026.
2. Claude API documentation, "Introducing Claude Fable 5 and Claude Mythos 5", accessed June 10, 2026.
3. The Decoder, "Anthropic releases Claude Fable 5 and Mythos 5 with major gains in coding and science", June 9, 2026.
4. TechCrunch, "Anthropic's Claude Fable 5 is a version of Mythos the public can access today", June 9, 2026.
5. AWS, "Anthropic Claude Fable 5 on AWS", June 9, 2026.
6. GitHub Changelog, "Claude Fable 5 is generally available for GitHub Copilot", June 9, 2026.
7. OpenRouter, model page for anthropic/claude-fable-5, accessed June 10, 2026.
8. Anthropic, "Redeploying Claude Fable 5", June 30, 2026, updated July 1, 2026.
9. Anthropic, Claude Fable model page (Fable 5.1, pricing), accessed October 16, 2026.