Claude Fable 5.2 and Mythos 5.2: faster, cheaper, and the same brain

On 9 November 2026, Anthropic unveiled on its website an update to its two most ambitious models, Claude Fable 5.2 and Claude Mythos 5.2. The company promises gains in coding, reasoning and vision, faster generation and a lower price.

Same engine, different keys

Fable and Mythos run on the same model. What sets them apart is the guardrails. Fable ships with reinforced safeguards around cybersecurity and biology, which lets Anthropic offer it to every user. Mythos is reserved for organisations doing beneficial work in those two fields, and they have to go through the Life Sciences and Cyber verification programmes first. Same car, same engine: Fable just gets the valet key.

Both versions cost $8 per million input tokens and $40 per million output tokens. Anthropic says that’s cheaper than the 5.1 generation, with generation up to 30% faster. An optional fast mode speeds up token output further, up to 2.5 times, for $16 and $80. Time is money, and here the exchange rate is exactly two to one.

Home-made benchmarks

According to Anthropic, both models sit a notch above Claude Opus 5.5 on almost every capability tested. The list covers agentic coding (FrontierCode 2.0), agentic search (WANDR), hardware design (BenchCAD 2.1), multidisciplinary reasoning (Humanity’s Last Exam) and chart reading (Chartography), which stands in for vision here.

On FrontierCode 2.0, Fable 5.2 at High effort reportedly roughly matches Opus 5.5 at Max effort, for about half the cost per task. At Xhigh effort, it would post the best score of any already-released model tested. At Medium effort, the default in the Claude apps and Claude Code, it would get through hard tasks using fewer tokens than Fable 5.1. These are the company’s own numbers, on tests it picked itself: the student marking their own homework, as tradition demands.

Code takes centre stage

Anthropic says more than 90% of the changes merged into its main codebase are now written by Claude, under human supervision. A team of agents running on a preliminary version of Fable 5.2 migrated a critical production service from Python to Rust. That service now starts about 26 times faster. The “let’s rewrite it in Rust” ticket that every team rolls over from one quarter to the next has finally found an assignee.

More eye-catching still: an engineer asked the model to design a prototype operating system, starting from an empty folder. Fable 5.2 coordinated more than 1,000 subagents, around the clock, for eight days. The result is an OS that boots on a laptop, with its own kernel to run the machine and drivers for the disk, keyboard and network. During testing, the company also says it got three working electronic devices designed from scratch, solutions to two open problems in black hole physics, and a formalisation of the proof of Fermat’s Last Theorem in 30 hours.

Safety and directions for use

Before release, Anthropic says it ran extensive safety testing, including new evaluations of worrying behaviours such as sandbox escape. The company judges that the risk of serious harm from misalignment remains low. Fable 5.2’s safeguards are said to be as robust as those of version 5.1, with fewer unwarranted refusals. Anthropic also claims its strongest protections to date against distillation attacks. The details are in the model’s system card.

Finally, Anthropic recommends starting with Claude Opus 5.5 for most coding and knowledge work, and moving up to Fable or Mythos only when the job really calls for it. Fable 5.2 can also act as an advisor to Haiku, Sonnet or Opus, or as an orchestrator that hands the simple tasks to smaller models.

So here’s a model that can write an operating system in eight days (Genesis built a whole world in less), and its maker advises you not to use it every day. It may be the first spec sheet to come with a “please enjoy responsibly” label.