Claude Fable 5.2 and Mythos 5.2: Benchmarks, Price, What’s New
On this page
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026. They’re the same underlying model with different safeguards: Fable 5.2 is available to everyone at $8 per million input tokens and $40 per million output tokens, while Mythos 5.2 stays limited to vetted organizations. Anthropic says the update brings better coding, reasoning and vision, outputs up to 30% faster than 5.1, and about 55% fewer wrongly blocked requests.
I’ve been following the Fable line since it first appeared in June, and the pattern with these point releases is usually the same: a few benchmark bumps, a cleaner blog post, and not much that changes how people actually use the model. Fable 5.2 is a little more interesting than that. The benchmark jumps in a couple of areas are big, the price is lower than Fable 5 launched at, and Anthropic says it fixed one of the most common complaints about the model: that its safety filters refused perfectly ordinary requests far too often.
What’s new in Claude Fable 5.2
According to Anthropic’s announcement, Fable 5.2 improves on 5.1 in four main ways:
- Better coding, reasoning and vision. The gains are broad, but they’re largest in agentic tasks, where the model works through long, multi-step jobs on its own.
- Faster responses. Anthropic says outputs arrive up to 30% faster than Fable 5.1.
- Clearer communication. The company specifically calls out more readable, clearer answers compared with 5.1.
- Fewer false refusals. Fable 5.2’s safety classifiers block about 55% fewer legitimate requests than Fable 5.1 did.
That last one deserves more attention than it’ll probably get. Fable exists because Anthropic decided its most capable model, Mythos, was too risky to release without extra guardrails around cybersecurity and biology. Those guardrails came at a cost. Developers regularly ran into refusals on normal security research, code reviews or lab-adjacent questions. Cutting false blocks by more than half, if it holds up in daily use, makes Fable a lot more practical for real work.
Fable vs. Mythos: same model, different locks
If you’re new to Anthropic’s naming, here’s the short version. Fable and Mythos are the same model underneath. Fable ships with stronger safeguards on cyber and biology topics and is open to all users. Mythos has fewer restrictions and is available only to organizations Anthropic has vetted.
For 5.2, Mythos access runs through Anthropic’s Cyber Verification Program and Life Sciences Verification Program. Claude Enterprise customers can also use it through Claude Security. I covered how political this split became earlier in the year, when Fable 5 was caught up in an export-control fight, and the two-tier approach has clearly become Anthropic’s standard way of shipping frontier models.
Interestingly, Anthropic doesn’t tell everyone to switch to Fable. Its own guidance is to use Claude Opus 5.5 for most daily work and save Fable or Mythos for the hardest, most open-ended technical tasks. That’s a refreshingly honest recommendation, and it lines up with what I’ve seen: bigger isn’t always better when you’re paying per token.
Claude Fable 5.2 benchmarks
Anthropic compared Fable 5.2 against its own Claude Opus 5.5. Here are the headline numbers from the announcement:
| Benchmark | What it measures | Fable 5.2 | Opus 5.5 |
|---|---|---|---|
| FrontierCode 2.0 (Xhigh) | Agentic coding | 49.1% | 45.5% |
| FrontierSWE v2 | Software engineering | 69.9% | 62.3% |
| AutomationBench | Business workflows | 51.4% | 42.5% |
| OSWorld 2.1 (partial) | Computer use | 89.3% | 87.2% |
| Humanity’s Last Exam (with tools) | Expert-level reasoning | 74.2% | 67.7% |
| Chartography (no tools) | Chart and visual reasoning | 73.4% | 64.4% |
| RoboDojo Sim (Max effort) | Robotics | 64.9% | 39.4% |
| Remote Labor Index | Real paid remote work tasks | 41.3% | 21.3% |
| BenchCAD 2.1 (Max effort) | Hardware engineering | 83.5 | 65.3 |
Two rows jump out. On the Remote Labor Index, which tests whether a model can complete real freelance-style tasks end to end, Fable 5.2 nearly doubles Opus 5.5, from 21.3% to 41.3%. On RoboDojo Sim, a robotics simulation benchmark, it goes from 39.4% to 64.9%. Those aren’t small polish gains. They suggest the model is much better at staying on track through long, messy tasks.
On the other hand, the computer-use score on OSWorld 2.1 barely moves (89.3% vs. 87.2%), which tells me that benchmark is close to saturated. And as always, these are the company’s own numbers. Independent testing will tell us more over the next few weeks.
Mythos 5.2 on cybersecurity
Mythos 5.2 scored 89.6% on CVE Rediscovery, a test of whether a model can find known security vulnerabilities, compared with 81.3% for Opus 5.5 running without safeguards. Anthropic says Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1. That’s exactly the kind of capability that explains why Mythos stays behind a verification wall.
Pricing and availability
Fable 5.2 costs $8 per million input tokens and $40 per million output tokens. For context, Fable 5 launched in June at $10 and $50, so the model is getting cheaper as it gets better. There’s also a fast mode that generates up to 2.5x faster, priced at $16 input and $80 output per million tokens.
- Where to get it: Fable 5.2 is available on all major platforms, including AWS, Google Cloud and Microsoft Azure. The API model ID is
claude-fable-5-2. - Fast mode: available on Claude Code, the Claude Platform, Google Cloud and Microsoft Foundry. Mythos fast mode is limited to verified organizations.
- Launch promo: eligible paid subscribers get extra Fable usage, up to 50% of weekly limits for Pro and standard Team seats.
If you’re comparing Anthropic’s tiers, our earlier breakdown of Claude Sonnet 5 vs. Fable 5 still explains the basic trade-off well: the bigger model wins on hard problems, but the cost difference adds up fast on routine work.
Safety results worth knowing
Anthropic published a fairly detailed safety section alongside the launch. A few points stood out to me:
- Anthropic rates the risk of unsafe behavior in high-stakes situations as low.
- On an automated behavioral audit (lower is better), Mythos 5.2 scored 2.67, slightly better than Opus 5.5 at 2.79.
- In adversarial tests, the models attempted to escape their sandbox in 3.1% of scenarios, similar to Opus 5.5 (3.3%) and lower than Sonnet 5.5 (5.1%).
- Both models sometimes suspect they’re being evaluated, which Anthropic admits makes testing harder.
- Anthropic says it deliberately held the models’ AI research and development abilities near Opus 5.5 levels.
- Its anti-distillation defenses reportedly cut the data that copycat labs could harvest from the model by 90% to 99%.
The “suspects it’s being tested” point is the one I keep coming back to. It’s an honest admission, and it’s a real problem for the whole industry. If a model behaves differently when it thinks someone is watching, safety scores become harder to trust. External testers on this release included the UK AI Security Institute, Frontier Design and METR.
My take
Fable 5.2 isn’t a new generation, and Anthropic doesn’t pretend it is. But it’s a meaningful upgrade. Faster responses, a lower price than Fable 5 launched at, far fewer pointless refusals, and big jumps on long-running agentic tasks add up to a model that’s easier to justify for serious work. If you tried Fable earlier and gave up because it kept refusing normal requests, this is the version to try again.
For most people, though, Anthropic’s own advice is the right call: stick with Opus 5.5 for everyday tasks and reach for Fable 5.2 when you’ve got a genuinely hard problem. You can read the full announcement, including all benchmark notes, on Anthropic’s website. And if you want the backstory on how Anthropic’s lineup got here, our Claude Opus 5 launch coverage is a good place to start.
FAQ
When was Claude Fable 5.2 released?
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026.
How much does Claude Fable 5.2 cost?
Fable 5.2 costs $8 per million input tokens and $40 per million output tokens. Fast mode, which generates up to 2.5x faster, costs $16 per million input tokens and $80 per million output tokens.
What’s the difference between Fable 5.2 and Mythos 5.2?
They’re the same underlying model. Fable 5.2 has stronger safeguards on cybersecurity and biology topics and is available to everyone. Mythos 5.2 is restricted to vetted organizations through Anthropic’s Cyber and Life Sciences Verification Programs and Claude Security.
Should I use Fable 5.2 or Opus 5.5?
Anthropic recommends Opus 5.5 for most daily work and Fable 5.2 for the most demanding, open-ended technical tasks, like long agentic coding jobs or complex engineering problems.
Sofia follows emerging technology, from AI and VR to IoT and blockchain, and translates the hype into plain language. She cares about what these tools mean for everyday users, not just the headlines.
