Anthropic dropped a big one today. On June 9, 2026, it released Claude Fable 5 — and it's calling it the most powerful model it has ever made generally available. Bold words. We build with these models every day, so the moment a new flagship lands, the first question on our minds isn't "what's the benchmark?" — it's "what does this actually change, and what does it cost?"
Here's the full picture: what Fable 5 is, the slightly confusing Mythos connection, the numbers Anthropic is claiming, the price tag, and the part nobody usually reads — the safety design. We'll keep the hype at arm's length and tell you what genuinely matters.
So what is Claude Fable 5?
Fable 5 is Anthropic's newest frontier model, and Anthropic describes it as a "Mythos-class model made safe for general use." That one sentence hides the whole story.
There are actually two models here: Mythos 5 and Fable 5. Mythos 5 is the full, less-restricted frontier model; Fable 5 is the version cleared for the public — the same core intelligence with stronger guardrails wrapped around it. As heise put it bluntly, Fable 5 is "Mythos 5 with restrictions." If you've followed Anthropic's "Mythos" research model in the rumor mill, this is the masses finally getting a (carefully gated) taste of it (VentureBeat).
The benchmarks — impressive, but read them as claims
Anthropic says Fable 5 is state-of-the-art on nearly all tested benchmarks (Tom's Hardware). The headline numbers it published:
| Benchmark (Anthropic-reported) | Fable 5 / Mythos 5 | For context |
|---|---|---|
| SWE-Bench Pro (real coding) | 80.3% | Opus 4.8: 69.2% · GPT-5.5: 58.6% · Gemini 3.1 Pro: 54.2% |
| Core long-task analytics | first to break 90% | ~10 points over Opus 4.8 |
| BioMysteryBench (hard) | 46.1% | Opus 4.8: 40.0% |
| ExploitBench | 78.0% | Mythos Preview: 69.0% |
A genuine, healthy dose of skepticism: these are vendor-reported numbers. Every lab presents the benchmarks where it shines, and a leaderboard score is not the same as your real workload. The honest move is to treat these as claims worth testing, not gospel — exactly the mindset we argued for in our ChatGPT vs Claude vs Gemini comparison. Independent evals over the next few weeks will tell the real story.
What it's genuinely good at
The most interesting line isn't a single score — it's the shape of the performance. Anthropic says the longer and more complex the task, the larger Fable 5's lead. That's a meaningful claim for anyone doing real work rather than one-shot chat: multi-step coding, long research, sprawling analysis.
The strengths cluster around software engineering, knowledge work, vision, and scientific research. That SWE-Bench Pro jump in particular is the one developers will care about, and it lines up with a trend we flagged this week — Anthropic itself reported that most of its own production code is now AI-written. The model that writes their code is the model they're now selling you.
The price — and what it actually means
Here's where builders sit up. Fable 5 (and Mythos 5) are priced at $10 per million input tokens and $50 per million output tokens (Anthropic).
That is frontier pricing — output is meaningfully more expensive than a workhorse model. For a heavy, automated app firing thousands of calls a day, that adds up fast. Which loops right back to the lesson we keep repeating: don't use a flagship for everything. Reach for Fable 5 on the genuinely hard, long, high-value tasks where its lead shows, and keep a cheaper model for the routine 80%. Our framework for choosing the right AI tool is built for exactly this kind of right-sizing — and it matters more, not less, as top-tier tokens get pricier.
How to actually try it
- Free, for a window: Fable 5 is included free for Claude Pro, Max, Team, and Enterprise users during an introductory period from June 9 to June 22, 2026 — a smart way to let people feel the difference before the meter starts.
- Developers: it's live on the Claude API as
claude-fable-5, and on Amazon Bedrock. - In the wild: specialized platforms are already onboarding it — legal-AI company Harvey announced availability the same day.
The safety design (the part worth noticing)
Most launch coverage skips this, but it's genuinely interesting. Because Fable 5 is the "made-safe" sibling of a more capable model, Anthropic bolted on an unusual safeguard: when Fable's classifiers detect a request touching cybersecurity, biology and chemistry, or model distillation, the response is automatically handled by Claude Opus 4.8 instead. Anthropic says this kicks in, on average, in under 5% of sessions.
It's a pragmatic, real-world admission: a more powerful model is also more dangerous in the wrong hands, so the riskiest 5% gets quietly downshifted to a tamer one. Whether that's elegant engineering or a usability speed bump depends on what you're building — security researchers and biotech folks may bump into the wall more than most.
Should you care? Our honest take
If you're a casual chat user, the practical change is modest — your Claude answers get sharper, especially on hard, multi-step questions, and during the free window you get it at no extra cost. Worth a try, not life-changing.
If you build — code, agents, research, analysis — this is the more interesting release of the season. The "lead grows with task length" framing, if it holds up independently, is exactly what production workloads need. This is also another data point in the bigger story we've been tracking: the AI race is increasingly about who owns the frontier model, and Anthropic just raised its own bar.
Our plan? Test it on the long, gnarly tasks first, measure the cost honestly, and keep a cheaper model for everything routine. That's not skepticism for its own sake — it's how you actually get value out of a tool this powerful (and this priced).
FAQ
What is Claude Fable 5? Claude Fable 5 is Anthropic's newest frontier AI model, released June 9, 2026, and described as a "Mythos-class model made safe for general use." Anthropic calls it its most powerful generally available model, with state-of-the-art results on most of the benchmarks it tested.
What's the difference between Fable 5 and Mythos 5? Mythos 5 is the fuller, less-restricted frontier model; Fable 5 is the version cleared for general public use, with stronger safety guardrails. In practical terms, Fable 5 is Mythos 5 with restrictions.
How much does Claude Fable 5 cost? Anthropic prices it at $10 per million input tokens and $50 per million output tokens — frontier-tier pricing. It's also free for Claude Pro, Max, Team, and Enterprise users during an introductory window from June 9 to June 22, 2026.
How do I access Claude Fable 5? It's available through the Claude API as claude-fable-5, on Amazon Bedrock, and free in the Claude apps for paid tiers during the intro window. Some third-party platforms (like Harvey) are also adding it.
Are the Fable 5 benchmarks reliable? The published scores are reported by Anthropic, so treat them as vendor claims until independent evaluations confirm them. Benchmarks also rarely match your specific workload — test on your own tasks before relying on the numbers.
The bottom line
Claude Fable 5 looks like a real step up — especially for developers and anyone running long, complex tasks — and the free introductory window makes it easy to judge for yourself. Just keep two things in mind: the benchmarks are Anthropic's until others verify them, and frontier pricing means you should use it where it earns its keep, not everywhere. Try it on something genuinely hard this week, watch the cost, and decide based on your work — not the launch-day headline.
