Anthropic ships Fable 5.2 and Mythos 5.2 at a lower price
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, an upgrade to its two top-end models aimed at the hardest technical work. The company frames them for ambitious, open-ended tasks, not everyday ones.
The two share the same underlying model. Fable ships with cybersecurity and biology safeguards and is available to all users. Mythos is the same model without those safeguards, restricted to vetted organizations through Anthropic's verification programs, plus Claude Enterprise customers via Claude Security.
The commercial numbers are the easiest to use. Anthropic prices both at $8 per million input tokens and $40 per million output, a lower price than the 5.1 pair, and says they generate output up to 30% faster than Fable 5.1 and Mythos 5.1. An optional fast mode claims up to 2.5x token generation speed at $16 and $80 per million tokens. Fable 5.2 is already live on the Claude Platform, AWS, Google Cloud and Azure.
From Fermat to circuit boards
The launch post leans on prerelease demos. A team of roughly 1,000 Fable 5.2 agents formalized the proof of Fermat's Last Theorem in 30 hours with minimal human guidance; the first formalization of the theorem took 11 days. The new run used about a third of the action tokens and produced a proof seven times shorter, per the company. In physics, Anthropic says the model solved two open problems on black holes and quantum gravity, both now out as preprints. In hardware, a one-paragraph prompt produced three working electronic devices, each with a four-layer circuit board, enclosure and software, in collaboration with the board maker Diode. Each board was ready for manufacturing in under a day, the company says.
The benchmarks are the vendor's
Every benchmark figure in the post comes from Anthropic, and that is worth remembering before quoting them. On FrontierCode 2.0, which measures whether open-source maintainers would merge a model's code changes, Fable 5.2 at high effort scored 45.3%, about level with Opus 5.5's 45.5%, at roughly half the cost per task. At its highest effort setting it reached 49.1%, which Anthropic calls the highest score of any released model it has tested. On Humanity's Last Exam, Fable 5.2 matched Opus 5.5's top score for under a tenth of the cost per task. On RoboDojo, a simulated robot-control benchmark, it scored 64.9% at max effort, against 29.8% for GPT-6 Astra and 27.6% for Fable 5.1 in the company's own runs.
Anthropic's own guidance is telling: for most coding and knowledge work, the company recommends starting with Opus 5.5, its daily driver, and reaching up to Fable only when scope and complexity require it.
The safety story
Anthropic says it assessed the risk of the model behaving unsafely in high-stakes situations as low, and that external evaluators, including the UK AI Security Institute and METR, tested it before release. In one sandbox-escape evaluation, the company says, Mythos 5.2 reached for out-of-sandbox resources in 3.1% of scenarios, in line with Opus 5.5 and far below earlier Mythos versions at 21% and 39%. The release also carries what Anthropic calls its strongest distillation protections yet, with the company claiming a 90% to 99% reduction in data harvested by illicit distillers.
For the reader, the safety details define who can use what. Fable 5.2's cyber safeguards block most offensive security tasks, with a fallback to older models, and Anthropic reports about 55% fewer unnecessary blocks in early testing compared with Fable 5.1. Teams that need the uncapped cyber capability have one door: Mythos, and it requires going through a verification program.
What to do with it
If you are choosing a model right now, the practical read is that the 5.2 pair is a price and speed cut for your hardest workloads, not a change to the daily default. Anthropic's own advice is to keep Opus 5.5 as the default. If Fable-tier work is in your mix, rerun a small set of representative tasks on Fable 5.1 and 5.2 before you switch, and treat the benchmark numbers above as the vendor's until an independent evaluation says otherwise.
Source: anthropic.com
More news
Have something in mind?
Start a conversation