Claude Fable 5 is here: Anthropic's first public Mythos-class model, with a safety wall built in
Anthropic launches Claude Fable 5 with a 1M-token context window, $10/$50 pricing, and a safety-classifier fallback — plus a restricted Mythos 5 for Project Glasswing partners.
Anthropic has shipped Claude Fable 5, its most powerful model to date and the first Mythos-class model available to the general public. At the same time, it quietly launched Claude Mythos 5 to a restricted group of cybersecurity partners. The two releases are, in a technical sense, the same underlying model. What separates them is a set of safety classifiers, and Anthropic’s confidence in who is on the other side of the API call.
What is a Mythos-class model, and why does the name matter?
Up until now, Opus was the top tier in Anthropic’s model family. Mythos sits above it. The naming is deliberate: “fable,” from the Latin fabula, is the Latin cousin of the Greek mythos. Fable 5 is the version made safe for general use. Mythos 5 is the same model with some of those safeguards lifted, available only to vetted partners.
This is a new structural approach for frontier AI. Rather than releasing one model and hoping misuse stays low, Anthropic has split a single capability level into two distinct products separated by a safety layer.
What Fable 5 can actually do
The benchmark numbers are significant. On SWE-bench Verified, the standard software engineering evaluation, Fable 5 scores 95.0%. Claude Opus 4.8 sits at 69.2%. GPT-5.5 scores 58.6%. On the harder SWE-bench Pro, Fable 5 reaches 80.3% against Opus 4.8’s 13.4%.
To put that in real terms: Stripe reported during early testing that Fable 5 completed a codebase-wide migration across a 50-million-line Ruby codebase in a single day. The same task, done by hand, would have taken a whole engineering team more than two months.
Vision capabilities have also taken a meaningful step forward. The model can rebuild a web app’s source code from screenshots alone. In a more unusual proof of capability, Anthropic says Fable 5 completed Pokémon FireRed with a minimal agentic harness. That is a long-horizon reasoning task with hundreds of sequential decisions. It is not a party trick.
Context window and pricing
Fable 5 ships with a 1 million token context window as standard, with up to 128,000 output tokens per request. Critically, there is no long-context surcharge. A 900,000-token request is billed at the same per-token rate as a 9,000-token one.
Pricing is $10 per million input tokens and $50 per million output tokens. Batch pricing drops to $5 and $25 respectively. Anthropic notes this is less than half the price of Claude Mythos Preview. The model IDs are claude-fable-5 and claude-mythos-5.
Adaptive thinking is always on. There is no way to disable it. The effort parameter controls thinking depth, but the raw chain of thought is never returned. This is a deliberate design choice, not a limitation of the API.
The safety classifier fallback
Here is the part that will matter most to developers building on top of the model. Fable 5 does not simply refuse high-risk queries. It routes them to Claude Opus 4.8 instead.
When that happens, the Messages API returns a stop_reason: "refusal" with an HTTP 200, not an error. The response includes a stop_details.category field that tells you which classifier fired: “cyber”, “bio”, “reasoning_extraction”, or null. Anthropic says this fallback triggers in fewer than 5% of sessions on average, and that the classifiers are tuned conservatively.
Billing follows the routing. If your prompt is routed to Opus 4.8, you pay Opus prices for the tokens processed there. If a request is blocked mid-conversation, the initial tokens are charged at Fable rates and the subsequent ones at Opus rates.
Where it is available, and for how long on subscriptions
Fable 5 is live on the Claude API, Claude Platform on AWS, Amazon Bedrock (US East N. Virginia and Europe Stockholm regions), Vertex AI, and Microsoft Foundry.
On subscriptions, the timeline is tight. Pro, Max, Team, and seat-based Enterprise plans include Fable 5 at no extra cost through 22 June. From 23 June, accessing the model requires usage credits. Anthropic has said it plans to bring Fable 5 back into subscription plans when capacity allows, but has not given a date.
One more change worth flagging: Anthropic is now requiring 30-day data retention on all Fable 5 and Mythos 5 traffic, including for enterprises that previously held zero-retention agreements. Anthropic says the data will not be used for training and will only be used to detect patterns of misuse across sessions. That is a reasonable justification, but if your organisation had zero-retention as a contractual requirement, this is worth checking with your legal team.
Claude Mythos 5 and Project Glasswing
Running alongside the public launch, Anthropic also upgraded its Project Glasswing partners to Claude Mythos 5. Project Glasswing began in April, giving Mythos Preview to around 150 organisations including financial institutions, healthcare networks, and critical software infrastructure providers. The restricted programme operates in collaboration with the US government.
Mythos 5 has safeguards lifted in certain areas, making it suitable for defensive cybersecurity work where the full capability of the model is needed. A separate biology programme will give selected life sciences researchers access to Fable 5 with biology and chemistry safeguards removed while cyber safeguards remain in place.
What this means for you
If you are a developer, the immediate priority is understanding the refusal routing before it surprises you in production. A 200 response with stop_reason: "refusal" is not an error, but it will behave differently to a normal completion. Build for it explicitly.
If you are on a subscription plan and want continued access to Fable 5 after 22 June, start calculating your likely usage now. Switching to usage credits mid-project is easier if you have planned for it.
If you are in enterprise software, the coding benchmark numbers are hard to ignore. The gap between Fable 5 and its nearest competitors on tasks like large-scale codebase migration is wide enough to have a material effect on engineering throughput.
And if your organisation had zero-retention agreements with Anthropic on Bedrock or the API, the new 30-day retention requirement needs a conversation with your compliance team before you migrate to Fable 5.
The model is available now. The access window on subscriptions closes 22 June.