On 9 November 2026, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, its newest AI models for demanding technical work. The launch looks like a routine point release on paper — a small version-number bump two months after Fable 5.1 arrived on 1 September 2026 — but the details matter: the new models are faster, cheaper, noticeably stronger at coding, and split into two versions with very different rules about who can use them.
If you use AI tools for coding, research, or long technical projects, this release probably affects you. Here is what actually changed, in plain English.
What is Claude Fable 5.2?
Claude Fable 5.2 is Anthropic's latest large language model, built for hard, open-ended technical work: long coding sessions, scientific research, and engineering projects that run for hours rather than minutes. It is the successor to Fable 5.1, and it is available now through the Claude API, the Claude apps, Claude Code, and the big cloud platforms — Google Cloud, Microsoft Foundry, and Amazon Web Services.
Anthropic describes Fable 5.2 as its most capable model for demanding technical work. That is a careful claim: for everyday questions and writing, Anthropic still points most users to Opus 5.5, its general-purpose flagship. Fable is the specialist — the model you reach for when a task is long, complex, and technical.
The model carries a one-million-token context window — roughly 750,000 words of text it can hold in mind at once — and can produce up to 128,000 tokens of output in a single response. Its training data runs to a June 2026 cutoff, which Anthropic lists as both its reliable knowledge cutoff and its training-data cutoff.
Faster and cheaper: the two headline changes
The most concrete improvements are speed and price, and both are easy to understand.
Up to 30% faster. Anthropic says Fable 5.2 generates output up to 30% faster than Fable 5.1. For a developer running a coding agent that writes thousands of lines, or a researcher asking the model to work through a long document, that speed difference is the difference between waiting and flowing.
20% cheaper. The API price has dropped from $10 to $8 per million input tokens and from $50 to $40 per million output tokens. In plain terms, a "token" is a small chunk of text — roughly three-quarters of a word — so a million input tokens is about 750,000 words read, and a million output tokens is about 750,000 words written. A 20% price cut on a model that is also faster is a genuine improvement, not a marketing repackage.
Anthropic also says Fable 5.2 at its "High effort" setting scores about as well on coding benchmarks as Opus 5.5 at its highest setting, at roughly half the cost per task. If that holds up in real use, it makes Fable 5.2 the better value choice for heavy coding work.
How much better is it at coding?
Benchmarks are the AI industry's scorecards, and they should always be read with a grain of salt — they are set by the companies themselves, and real-world performance varies. With that caveat, Anthropic's published numbers show a clear jump:
| Benchmark | Fable 5.1 | Fable 5.2 | Opus 5.5 |
|---|---|---|---|
| FrontierCode 2.0 (coding) | 32.4% | 49.1% | 45.5% |
| CVE Rediscovery (finding known vulnerabilities) | 78.3% (Mythos 5.1) | 89.6% (Mythos 5.2) | 81.3% |
The FrontierCode 2.0 result is the striking one: Fable 5.2 jumps from 32.4% to 49.1%, overtaking Opus 5.5's 45.5%. That is a large gain for a point release, and it is the main reason developers are paying attention.
Anthropic also offered a more dramatic demonstration: a team of about 1,000 Fable 5.2 agents worked together to formalise the proof of Fermat's Last Theorem — one of mathematics' most famous problems — in the Lean proof language, in about 30 hours. The company also says the model helped solve two open problems in the physics of black holes, with preprints posted for both results. These are Anthropic's claims, not independent measurements, but they show what the model is being aimed at: long, multi-step technical work, not quick chat answers.
Fable vs Mythos: one model, two doors
Here is the part most people will find confusing, so let us be precise: Fable 5.2 and Mythos 5.2 are the same underlying model. Anthropic says so plainly. The difference is not what the model can do — it is what it is allowed to do, and who is allowed to use it.
Fable 5.2 is the broadly available version. It comes with strong safeguards around cybersecurity and biology — the two areas where a powerful AI could be misused — and that is what allows Anthropic to offer it to everyone: paid Claude plans (Pro, Max, Ultra, Team, and Enterprise), Claude Code, and the cloud platforms. Developers can start using it with the model ID claude-fable-5-2.
Mythos 5.2 is the same model with some of those safeguards relaxed, for work that legitimately needs them — cybersecurity defence research and life-sciences work, for example. Access is limited to vetted organisations through Anthropic's Cyber Verification Program and Life Sciences Verification Program. Both versions cost the same ($8 in, $40 out per million tokens), and they share the same specifications.
Think of it like a laboratory: Fable is the public lab with strict safety rules, Mythos is the same lab with a restricted wing for qualified researchers. The 89.6% vulnerability-discovery score quoted above, for instance, is a Mythos result — it should not be treated as a promise that the public Fable version will answer the same requests.
What else changed under the hood?
Beyond speed and price, Anthropic's developer documentation flags a few changes worth knowing before you upgrade:
- Always-on adaptive thinking. The model now decides for itself how much reasoning a task needs, rather than waiting to be told.
- Restrictions on forced tool calls. Developers can no longer force the model to call a specific tool in some situations, which changes how some automated workflows behave.
- 30-day data retention. Most customers now face a 30-day retention period for safety monitoring, though customers eligible for zero data retention can keep those arrangements.
Anthropic also says Fable 5.2 has its strongest protections yet against distillation attacks — attempts to copy a model's abilities by training a cheaper model on its outputs.
The bigger picture: an IPO is near
The launch lands in a delicate week for Anthropic. CNBC has described it as the company's first frontier release since chief executive Dario Amodei called in September for the AI industry to slow the pace of capability gains — and reported that Anthropic's initial public offering could happen as soon as next week, with a targeted valuation of $1 trillion to $1.5 trillion, down from a floated $2 trillion. Those IPO figures are reports, not confirmed by the company, but they explain why this release is being watched so closely: a faster, cheaper, stronger model is a good story to tell investors.
There is also a wider industry backdrop. The same week, chipmaker TSMC approved plans to invest up to $500 million each in Anthropic and OpenAI, a sign of how deeply the AI labs and the chip supply chain are now tied together.
Should you switch?
For most everyday users, the honest answer is: probably not yet, and Anthropic itself agrees — it still recommends Opus 5.5 for general work. Fable 5.2 earns its place when your work is long, technical, and code-heavy: a multi-file coding project, a research task that runs for hours, or an engineering problem with many moving parts.
If you are a developer, the switch is cheap to try: the price is lower, the speed is higher, and the model ID is claude-fable-5-2. Run it on one real task before committing — benchmarks are Anthropic's, not yours. And if your work touches cybersecurity or biology, remember that the version you can use (Fable) is deliberately not the version that scored 89.6% on vulnerability discovery (Mythos). That gap is not a bug. It is the whole point of having two doors to the same model.
Takeaway checklist
- Fable 5.2 launched 9 November 2026 with Mythos 5.2 — the same model, two access levels.
- Up to 30% faster output than Fable 5.1, and 20% cheaper: $8 in / $40 out per million tokens.
- Coding jump: 49.1% on FrontierCode 2.0, up from 32.4% — ahead of Opus 5.5's 45.5%.
- 1M-token context window, up to 128,000 output tokens, June 2026 knowledge cutoff.
- Fable is for everyone (Claude plans, API, Google Cloud, Microsoft Foundry, AWS); Mythos is for vetted organisations only.
- For everyday use, Opus 5.5 is still Anthropic's recommendation — Fable 5.2 is the specialist for long technical work.
- Try it on one real task before switching your workflow — benchmarks are the maker's own numbers.
Read next: Large Language Models Explained and How AI Agents Are Changing Work.
