Summary
- Claude Pro subscribers and standard Team seats can use Fable 5.2 at no extra cost through November 16.
- Fable 5.2 is up to 30% faster and costs 20% less per token than Fable 5.1.
- Anthropic says Fable 5.2 offers major improvements in coding, robotics, efficiency, and safeguards.
Anthropic has launched Fable 5.2, and Claude Pro subscribers and standard Team seats can use it at no extra cost through November 16 at 11:59 p.m. Pacific Time. Fable can consume up to 50% of their existing weekly allowance during that window. Eligible accounts receive one Fable reset during the promotion. After that, it returns to pay-as-you-go usage credits on those plans.
While Fable 5.1 was already pretty impressive, Fable 5.2’s announcement comes with some mind-boggling stats, achievements, and testimonials. The model managed to coordinate over 1,000 AI agents to build a functional operating system from scratch, and it took only eight days. It also formalized the proof of Fermat’s Last Theorem in 30 hours and helped solve two open problems involving black holes.
Fable 5.2 spends fewer tokens and accomplishes more
Anthropic’s benchmark table shows some sizable jumps over Fable 5.1. Let’s take FrontierCode 2.0, which judges whether open-source maintainers would merge an AI’s code, as an example. On that benchmark, Fable 5.2 managed to perform more than 30% better than its predecessor. Moreover, in simulated robotics, the score more than doubled by jumping from 27.6% to 64.9%. In practical terms, the model appears considerably better at writing production-ready code and performing complex tasks like controlling robots from camera input.
Fable 5.2 also generates output up to 30% faster and charges $8 per million input tokens and $40 per million output tokens, down from Fable 5.1’s $10 and $50, respectively. More importantly, Anthropic claims it can solve difficult problems more efficiently and use fewer tokens to achieve objectives. At High effort, it roughly matched Opus 5.5 at Max effort on FrontierCode while costing about half as much per completed task.
Anthropic is keeping a tighter leash on this release
With all the recent news about the dangers of artificial intelligence, it only makes sense that AI companies are trying to showcase security awareness as much as they show excitement about their new models’ capabilities. AI agents have already caused a lot of trouble by taking actions their creators didn’t intend, including an OpenAI agent that breached an Australian government website. Famously, a former Anthropic researcher even left the industry warning that leading labs were racing toward systems that could potentially end humanity very soon. Fable 5 itself also caused a lot of controversy upon its release. Only three days after it rolled out publicly, Anthropic had to shut it down as the US government raised concerns about a method for bypassing its cybersecurity safeguards.
Anthropic says Fable 5.2 and Mythos 5.2 were tested on more than 1,800 adversarial sandbox scenarios. The underlying model managed to access files outside its sandbox in 3.1% of the cases. It’s not 0%, but compared with 21% for Mythos 5.1, it seems that the new models have substantially better safeguards.
