Claude Fable 5 Is Already Here: Where It Runs, the Model ID, and What It Really Costs
Claude Fable 5 isn't coming — it shipped June 9, 2026 and is available now via the Claude API, Amazon Bedrock, Google Vertex AI, Microsoft Foundry, and claude.ai. The model ID is claude-fable-5. It costs $10 per million input tokens and $50 per million output, with a 1M-token context window.
Update — June 30, 2026: Claude Fable 5 (
claude-fable-5) is back. After the June 12 US export-control directive that had taken both models offline was lifted on June 30, Anthropic is redeploying Fable 5 globally on July 1, 2026 across the Claude Platform, Claude.ai, Claude Code, and Claude Cowork (cloud providers to follow). It returns with new safety classifiers that block certain harmful cybersecurity tasks; on Anthropic’s own surfaces a blocked request falls back to Opus 4.8, and Anthropic flagged more false positives in routine coding and debugging that it will keep refining. See what changed in the redeployment and how to handle the classifier fallback as a developer. The analysis below describes Fable 5 as it launched; treat its availability and pricing as current again.
The short answer: it already shipped (June 9, 2026)
If you typed “when will Claude Fable 5 be available” into a search box, I have news: the wait is already over. Anthropic made Claude Fable 5 generally available on June 9, 2026 — about two weeks ago as I’m writing this. The official line in Anthropic’s announcement is that it’s “available everywhere today.”
One thing trips people up. Fable 5 launched alongside Claude Mythos 5, and they are not the same situation. Mythos 5 is not general — it’s limited to “Project Glasswing” approved partners and select biology researchers, and for those authorized users it ships with lifted cybersecurity guardrails. Fable 5 is the safeguarded version of Mythos, made safe for general use. Anthropic frames Mythos-class models as a new tier that sits above the Opus class in capability — so Fable 5 is the public, guardrailed slice of something more powerful.
My take, flagged as opinion: the real question stopped being “when” the moment it hit GA. The useful question now is “where do I run it, and can I afford to?” That’s the rest of this post.
Where can you run Claude Fable 5, and what’s the model ID?
You can reach Fable 5 on five surfaces: the Claude API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, and claude.ai. Anthropic’s phrasing was “available everywhere today,” and for once that’s close to literal.
If you’re writing code, the model ID you target is claude-fable-5. Inputs are text, image, and file. Output is text, with reasoning / extended thinking support. The context window is 1,000,000 tokens, which is the part that changes how you architect long-horizon work — you can keep a lot of state in the prompt instead of bolting on retrieval for everything.
Claude Fable 5 at a glance
What does Claude Fable 5 actually cost?
Here are the hard numbers: $10 per million input tokens and $50 per million output tokens. That’s roughly 2x Claude Opus 4.8, which sits at $5 input / $25 output. Anthropic also notes it’s “less than half the price of the Claude Mythos Preview,” so depending on what you’re comparing against, the framing flips.
The lever that matters most is prompt caching. Anthropic gives a 90% discount on cached input, which drops your stable prompt prefix to about $1 per million cached input tokens. If you run agents with a big, mostly-static system prompt, that’s not a rounding error — it’s the difference between viable and painful.
Be honest about where the bill actually lives, though: at $50/M, output is the expensive side. Output length and your caching strategy move your invoice far more than raw input volume does. I dig into this math in detail in how much it costs to add AI to your app.
Cost per 1M output tokens
The access timeline: the free window on subscriptions just closed
This one has a sharp edge, so let me be precise. On claude.ai paid plans — Pro, Max, Team, and seat-based Enterprise — Fable 5 was free, at no extra cost, through June 22, 2026.
Starting June 23, 2026 (which, as I publish this, is today), it requires usage credits “until capacity allows standard inclusion.” So the free ride just ended. On the API and consumption-based Enterprise, none of this applied — it was fully pay-as-you-go from day one, no free window.
Honesty note: if you tried Fable 5 for free last week on Pro or Max and it feels different today, that’s not a glitch. That’s the policy change kicking in, right on schedule.
How do I start using it today?
Start using Claude Fable 5 today
- Pick your surfaceclaude.ai for chat; Claude API / Bedrock / Vertex AI / Foundry for code
- Target the model IDIn code, call claude-fable-5
- Send your inputstext, image, or file — use the 1M-token context for long-horizon work
- Turn on prompt cachingGet the ~$1/M cached-input rate on your stable prompt prefix
- Mind the billingOn a paid claude.ai plan, expect to spend usage credits now; on the API, you're billed pay-as-you-go
That’s genuinely it. The 1M-token context is the part worth designing around — it changes what “one task” can mean. If you’re standing up something durable, the patterns I lean on (routing, caching, fallbacks, cost caps) live in putting an LLM in production.
A builder’s cost playbook: when to reach for Fable 5 (and when not to)
The discipline here is simple: route by task complexity. Reserve Fable 5 for ambitious, long-horizon work where its autonomy actually pays off. Anthropic says it “can work autonomously for longer than any previous Claude models” and “stays focused across millions of tokens in long-running tasks.”
The example from Anthropic sells it well: it says Stripe used Fable 5 to do a codebase-wide migration on a 50-million-line Ruby codebase in one day — work it says would have taken a team over two months by hand. That’s the shape of job where $50/M output is a bargain.
For everything else — short, cheap, high-volume calls — drop down to a cheaper model like Opus or Haiku. Paying $50/M output for a one-line classification is just waste. Plenty of work doesn’t need a frontier model at all, which I cover in AI agents vs automation tools.
There’s a billing nuance worth internalizing: Fable 5’s classifiers detect cybersecurity, biology/chemistry, and model-distillation topics, and when they fire, the response is automatically handled by Opus 4.8 instead (and you’re notified). So on those specific tasks you effectively get Opus 4.8 — meaning you’re paying Fable rates for Opus-level output there. Anthropic says more than 95% of Fable sessions involve no fallback, so for most workloads this won’t bite, but it’s real.
My take, as opinion: don’t default everything to Fable 5. The headline benchmarks are strong — SWE-Bench Pro 80.3%, spatial reasoning around 3x Opus 4.8, leading legal reasoning — but read the fine print: some of the top benchmark rows reflect Mythos 5, not Fable 5, and on cybersecurity, biology, and distillation tasks Fable 5’s safeguards make it fall back to Opus 4.8, so its real-world performance there is effectively Opus 4.8. Calibrate expectations to the model you can actually call.
When to reach for Fable 5 (and when not to)
Use Fable 5 when
- Ambitious, long-horizon, agentic work
- Large-context migrations (Anthropic cites Stripe: 50M-line Ruby in a day)
- Autonomy across millions of tokens pays off
- Hard reasoning where SWE-Bench Pro 80.3% shines
Reach for cheaper when
- Short, high-volume, simple calls
- One-line classifications and routing
- Cost-sensitive batch work — drop to Opus or Haiku
- Tasks that hit the Opus 4.8 fallback anyway
Two things worth knowing before you commit
I’m not here to dunk on a strong model, but two things deserve a clear head before you route client traffic through it.
First: a mandatory 30-day data-retention policy applies to all Fable/Mythos traffic — even customers with zero-retention agreements. Anthropic frames this as needed to defend against novel attacks. Reporting notes it could set a precedent where access to powerful models comes bundled with mandatory retention justified as “safety.” If you handle sensitive data, that’s a real consideration, not a footnote.
Second, the timing. Anthropic released Fable 5 publicly just days after warning that AI is becoming too dangerous — it had urged major labs to adopt “a coordinated brake pedal on frontier AI development,” cautioning that systems may soon reach recursive self-improvement (improving themselves without human oversight). Critics read the release as a contradiction between stated caution and aggressive commercialization. I’ll let you weigh that one yourself.
On security specifically, Anthropic reports no universal jailbreaks in over 1,000 hours of external testing, while acknowledging “there could still be novel attacks.” That’s a fair, non-absolute claim, and I respect that they didn’t overstate it.
My take, as opinion: the capability is genuinely strong. But the mandatory 30-day retention is the part I’d weigh hardest for client work with strict data terms — it’s the kind of clause that can quietly disqualify a model regardless of how good the benchmarks look.
FAQ: Claude Fable 5 access and cost
When will Claude Fable 5 be available? It already is — generally available since June 9, 2026.
What’s the model ID? claude-fable-5.
How much does Claude Fable 5 cost? $10/M input, $50/M output (~2x Claude Opus 4.8), about $1/M for cached input, with a 1M-token context window.
Is Claude Fable 5 free on Pro/Max/Team/Enterprise? It was, through June 22, 2026. From June 23 it requires usage credits “until capacity allows standard inclusion.”
Where can I run it? Claude API, Amazon Bedrock, Google Vertex AI, Microsoft Foundry, and claude.ai.
How is it different from Claude Mythos 5? Mythos 5 isn’t general — it’s limited to Project Glasswing partners and select researchers, with lifted cybersecurity guardrails for authorized users. Fable 5 is the safeguarded, generally available version.
The bottom line
Fable 5 isn’t a “coming soon.” It’s here, it’s pay-as-you-go on the API, and the subscription free window closed on June 23. Route deliberately: Fable 5 for long-horizon, ambitious work; cheaper models and aggressive caching for everything else. And before you push client traffic through it, weigh the mandatory 30-day retention against your data terms — for some engagements that’s the deciding factor, full stop. If you want the deeper playbooks, my AI engineering guides collect the cost math and production patterns in one place.