Testing the new Claude Fable 5 model in real-world conditions: Does it live up to expectations?

The Claude Fable 5 is the first model in Anthropic's Mythos line to be released to the market, a level previously locked down after Project Glasswing because its cybersecurity capabilities were deemed too robust for widespread release.

Fable gives the feeling of a different model.

Why Fable 5's reasoning and writing style stand out from the very first glance.

The numbers surrounding Fable 5 are just as impressive as the user experience. It outperforms Claude Opus 4.8 by over 10% in several performance tests, spreadsheet tasks complete 25 to 30% faster, and it's the first Anthropic model capable of building an entire application in a single run. Legal teams in the anonymous review process stated it consistently achieved performance comparable to or exceeding their existing model. Researchers described it as being at the senior research scientist level.

But benchmarks are just benchmarks, and the real-world use of an AI model can be vastly different. Once given access to Fable 5, many people have tried using it to fix Linux kernel bugs, 3D printers that aren't fully calibrated, get advice on an unfinished ESP32 project, and sometimes, choose an outfit to wear to a family gathering.

Claude Fable 5 remains calm even when things go wrong.

Debugging, troubleshooting, and handling complex issues without disruption.

The author's week working with Fable started off badly. The author was trying to install SnapOtter—a self-hosted photo editing application as an alternative to cloud-based photo editing apps—on their Linux Mint machine with GPU acceleration. It turned out that an Nvidia driver needed to be installed, but a faulty driver installation corrupted the boot sequence on the latest kernel, and attempts to fix the problem resulted in a kernel panic. At this point, the machine running all self-hosted applications became completely unbootable.

Fable 5 accurately identified that initramfs on one of the older kernels was not corrupted during the driver rebuild process — no GRUB, no Secure Boot, no other complex issues commonly encountered. And it did all of this with just simple messages and images from the Linux terminal captured from the phone. It restored the system using the old operating system kernel and eliminated the driver installation error.

During the uninstallation process, it also discovered that running autoremove had removed linux-modules-extra-6.17.0-35-generic, the package containing the iwlwifi driver, meaning the machine lost Wi-Fi on the already fixed kernel. The workaround was to pin the image, modules, and modules-extra along with linux-generic-hwe-24.04. Resolving such chain issues with older models is quite challenging. Fable 5 saw the whole picture and helped address the core problem instead of just fixing the first bug it found.

Another instance where Fable 5 helped solve a problem was when trying to run Shadowbroker inside Docker but being unable to access it from another machine on the network. The model tried various settings such as port bindings, UFW, iptables, ss output, and more. The solution turned out to be embarrassingly simple: The author had mistakenly used the IP address of the Linux machine for the IP address of the Home Assistant virtual machine during the session. Fable 5 realized the author was using the wrong IP address by querying old conversations that needed to access something running on their Linux machine using another device, and pointed out exactly what needed fixing.

Claude Fable 5 is not just a thinking tool - it can also build

Fable has demonstrated its usefulness in various projects, code snippets, and workflows.

Fable 5 can help you do more than just repair faulty operating systems and Docker containers. The author also used the model's help to figure out why the Bambu Lab A1 Mini 3D printer was experiencing filament pulling and weak printing on Numakers' new PLA+ filament—something never done before. It prompted the author to print a temperature tower to find the nozzle temperature, stabilize it at the appropriate temperature, and help calculate the pressure rise values.

Another print kept falling off the print bed midway, in addition to geometric issues. Fable 5 discovered that the 3MF file being used had a built-in generic ABS profile, which was completely ignored. Importing that file brought the print bed temperature to 90°C and turned off the cooling, which ruined the PLA+ print. Now that Claude can design 3D parts for you, Fable 5 also helped the author design a custom camera mount for his motorcycle — something older Anthropic models didn't do well.

The most productive conversation of the week with Fable 5 began with a simple question about components. The author wanted a three-step potentiometer with a click sound for a project they were planning. Fable 5 corrected the question, stating that an on-off-on switch was needed. From there, the conversation expanded into a complete electronics project. Fable 5 was able to find the answers to all the goals being attempted, including retrieving information using Claude Code and using the ESP32 to send temperature and humidity data to the Home Assistant system via the DHT22 sensor.

The author also asked Fable 5 for clothing advice, just to see how it would respond. The author needed to consider whether a black or light blue striped shirt would be more appropriate for a family gathering, as the author only had beige shoes and a belt. The app suggested a light blue striped shirt to keep the outfit harmonious, and it was correct.

The most important strengths are not shown on the ranking.

Reliability, judgment, and qualities that are difficult to measure by standards.

The attention given to Fable 5 is focused on aspects such as cybersecurity, molecular biology, and long-range automated programming. That's where it seems unprecedented. But perhaps more importantly, it's far more coherent—at least within a typical, messy, multidisciplinary work week. It maintains context during long sessions, avoids complicating things when the answers are simple, and operates within real-world constraints rather than imaginary ones.

It's not that you can't do these things with an older model. You absolutely can, but the amount of effort and time required to achieve the same answer would be much higher. Most models become less accurate as the work session gets longer and more complex. Fable 5 doesn't—and that's more important than any standard. Access to Fable 5 is currently suspended while Anthropic addresses the export control issue, but if it is restored, there will undoubtedly be a large backlog of requests.