Testing Anthropic’s Claude Fable 5: Two Queries Cost 90 CNY

Claude Fable 5 is the public‑facing version of Anthropic’s model. It offers strong capabilities but applies stricter safety filters; for security‑sensitive, biochemical, or model‑extraction queries it falls back to the weaker Opus 4.8.

Mythos 5 is a limited‑access variant that shares the same core engine as Fable 5 but relaxes some restrictions for trusted institutions, such as allowing more powerful security‑analysis features.

After upgrading the Microsoft Copilot CLI, the model list showed Fable 5. Its listed price is roughly twice that of Opus 4.7, making it noticeably more expensive than the $39 Copilot Pro plan.

Capability test – fireworks prompt

Using the maximum “think‑depth” setting (Max) and a 1 000 000‑token context window, a classic fireworks description was sent. The model generated vivid, physically plausible fireworks: spherical peonies, meteor‑rain with realistic trailing lines, heart shapes, tilted halos, hanging golden willows, and bi‑color hydrangeas. The physics included gravity‑driven parabolic trajectories, air‑resistance damping, brightness decay, and random flicker.

The same prompt was run on the leading Chinese model Qwen 3.7‑Max for comparison; the entire test cost about $3.

Optimizing a personal knowledge‑collection tool

The “Skills” tool crawls web pages, follows links up to five layers deep, and saves images locally. Before optimization it only captured main text and often saved thumbnail images instead of full‑size ones; video download was not supported.

Claude Fable 5 was prompted to learn the tool’s workflow and to add video‑download capability. Over a 20‑minute interactive session the model identified the thumbnail‑saving issue, modified the code to retrieve full‑resolution images, and added logic to download embedded videos.

The optimization session cost roughly $10 (≈ 90 CNY). After the changes, the tool successfully saved previously broken images and could download videos, confirming the new functionality.

Cost considerations

Both the fireworks benchmark ($3) and the Skills‑tool optimization ($10) illustrate that Fable 5’s usage incurs a higher monetary cost than comparable domestic models, despite its strong generative ability.

Project repository:

github.com/tjxj/z-skills