Claude Fable 5.2 benchmarks: what they mean for calls
On 9 November 2026 Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, its most capable models for demanding technical work. They are the same model in two versions: Fable is the public one, while Mythos has fewer cybersecurity and biology safeguards and is restricted to vetted organizations. For an AI phone assistant nothing changes overnight: the official announcement says nothing about voice and focuses on coding, finance and science. Still, it tells you where the models behind AI agents are heading.
What shipped, with Anthropic's numbers
- Availability. Fable 5.2 is available through the API as
claude-fable-5-2, in the Claude apps, and on Amazon Web Services, Google Cloud and Microsoft Azure. Mythos 5.2 is only open to participants in the cybersecurity and life-sciences verification programs. - Pricing. $8 per million input tokens and $40 per million output tokens. Fast mode costs $16 and $80, and generates up to 2.5x faster.
- Speed. Anthropic says outputs are up to 30% faster than Fable 5.1 and Mythos 5.1, and the model uses fewer tokens at the same effort.
- Benchmarks. Fable 5.2 beats Fable 5.1 in every published test (table below): for example 49.1% vs 32.4% on FrontierCode 2.0.
- Safety. Fable 5.2 blocks about 55% fewer requests than 5.1. Anthropic rates the risk of misaligned behavior as low.
One detail worth noticing: Anthropic recommends Opus 5.5 as the everyday model. Fable 5.2 is meant for the hardest tasks, including as an "advisor" or "orchestrator" for smaller models that do the simple work.
Fable 5.2 benchmarks
Anthropic publishes Fable 5.2's results against Fable 5.1 and Opus 5.5, and for some tests also against GPT-6 Astra and Gemini 4 Carbon. These are figures reported by Anthropic in the announcement, not independent measurements.
| Benchmark | Fable 5.2 | Fable 5.1 | Opus 5.5 | Other models |
|---|---|---|---|---|
| FrontierCode 2.0 (agentic coding) | 49.1% | 32.4% | 45.5% | GPT-6 Astra 42.8%, Gemini 4 Carbon 41.1% |
| FrontierSWE v2 | 69.9% | 56.3% | 62.3% | GPT-6 Astra 65.5%, Gemini 4 Carbon 61.2% |
| AutomationBench | 51.4% | 31.4% | 42.5% | GPT-6 Astra 41.4%, Gemini 4 Carbon 53.9% |
| Terminal-Bench-Science 0.1 | 74.6% | 52.6% | 58.7% | Mythos 5.2 75.3% |
| WANDR (relaxed) | 82.1% | 68.7% | 72.3% | n/a |
| OSWorld 2.1 (offline, partial) | 89.3% | 83.5% | 87.2% | n/a |
| Remote Labor Index | 41.3% | 17.9% | 21.3% | n/a |
| Humanity's Last Exam (with tools) | 74.2% | 65.6% | 67.7% | n/a |
| Chartography (no tools) | 73.4% | 44.8% | 64.4% | n/a |
How to read them:
- The jump over 5.1 is clear in every test listed, for example +16.7 points on FrontierCode 2.0 and +20 points on AutomationBench.
- Against Opus 5.5 the lead is smaller, from 2.1 points on OSWorld 2.1 to 20 points on Remote Labor Index. That is why Anthropic still points to Opus 5.5 for everyday use.
- It does not win everywhere. On AutomationBench, per Anthropic, Gemini 4 Carbon (53.9%) beats Fable 5.2 (51.4%). We covered that model in Gemini 4 Carbon.
- No benchmark measures a phone call. These are tests of coding, office-task automation and science: they say nothing about how natural a voice sounds or how short the pause before an answer is.
AutomationBench and OSWorld are the closest to what an assistant does during a call, using tools and software to finish a task. Even there, though, the tasks are long ones, not a three-minute booking.
What it changes for a call handled by AI
A call with an AI assistant has three steps: transcribing what the customer says, reasoning, and a voice that answers. Fable 5.2 only concerns the second, and its strength is long, complex work rather than a three-minute call to book a table.
- The biggest model does not win on the phone. On a call every fraction of a second is audible: to book a visit, a light, fast model often works better than a flagship one. We saw the same with Claude Haiku 5.5.
- The orchestrator idea is interesting. A large model that decides and small models that execute suits an assistant that checks a calendar or a practice-management system during the call. It is a possibility, not a feature anyone has announced for voice.
- Dollars per token are not your costs. API prices concern the service provider. What matters to you is the assistant's subscription: see the pricing page.
- No news on voice. To hear a difference on the phone, speech and synthesis models matter, not this announcement.
Who should care most
- Professional firms: files, appointments and requests to route. More reliable reasoning helps with articulated requests, but answering immediately stays the priority.
- Medical practices: promptness and data handling matter more than any coding benchmark.
- Real estate agencies: qualifying a lead (area, budget, timing) needs good reasoning, at call speed.
What to do now
Nothing urgent. The model is a component the provider updates behind the scenes, and the right choice is not "the newest model" but the one that works best on your calls. If you are evaluating an AI phone assistant, ask:
- How long between the end of the customer's sentence and the start of the answer?
- How does it handle spoken language and accents?
- What actions does it take during the call (calendar, WhatsApp, handoff to a person)?
The fastest way to find out is to hear it: try the demo and judge from a real call.
Frequently asked questions
What are the Claude Fable 5.2 benchmarks?
According to Anthropic, Fable 5.2 scores 49.1% on FrontierCode 2.0, 69.9% on FrontierSWE v2, 51.4% on AutomationBench and 89.3% on OSWorld 2.1, always above Fable 5.1. On AutomationBench Gemini 4 Carbon scores 53.9%.
What is Claude Fable 5.2?
It is Anthropic's most capable publicly available model, released on 9 November 2026. It is available through the API as claude-fable-5-2, in the Claude apps, and on Amazon, Google and Microsoft cloud platforms.
What is the difference between Fable 5.2 and Mythos 5.2?
According to Anthropic they are the same underlying model. Mythos has fewer cybersecurity and biology safeguards and is limited to vetted organizations; Fable is the one open to everyone.
Is Fable 5.2 a voice model?
No. The announcement does not mention voice or phone response latency: it is about technical work, coding, finance and science. A phone call still needs a system that combines transcription, reasoning and speech synthesis.
Will my AI phone assistant use Fable 5.2?
Not necessarily. Anthropic itself recommends Opus 5.5 as the everyday model and Fable 5.2 for the hardest tasks. For a phone assistant, response time, spoken language handling and reliable actions matter most.
