Can AI detectors catch Claude Fable 5.1?
Mostly not. Claude Fable 5.1 was the hardest of the four Claude models to detect. A majority of the seven free detectors we tested flagged only 38.9% of its texts. Fast-DetectGPT caught 46.0% and Desklib 32.5%, so most Fable texts passed as human on the two most reliable detectors.
Tested
A majority of the detectors (4 or more of 7) flagged49 of 126 Claude Fable 5.1 texts (38.9%).
Numbers behind this chart
| Item | Rate | Texts | 95% interval |
|---|---|---|---|
| TMR (Target Mining RoBERTa) | 88.1% | 111 of 126 | 81.3 to 92.7% |
| Fakespot AI text detector | 50.0% | 63 of 126 | 41.4 to 58.6% |
| E5-small LoRA detector | 49.2% | 62 of 126 | 40.6 to 57.8% |
| Fast-DetectGPT (Qwen2.5-1.5B) | 46.0% | 58 of 126 | 37.6 to 54.7% |
| Desklib AI text detector | 32.5% | 41 of 126 | 25.0 to 41.1% |
| MAGE | 28.6% | 36 of 126 | 21.4 to 37.0% |
| OpenAI RoBERTa detector | 3.2% | 4 of 126 | 1.2 to 7.9% |
How we tested Fable
Claude Fable 5.1 answered the same 150 prompts as the other models, matched to the topic, genre and length of the human texts. The middle Fable text was 274 words long. We report on the 126 texts outside the calibration split.
Results
Fable was the hardest model for almost every detector. Fakespot caught 50.0%, E5-small LoRA 49.2%, Fast-DetectGPT 46.0%, Desklib 32.5% and MAGE 28.6%. The OpenAI RoBERTa detector caught 3.2%.
TMR is the exception, with 88.1%, its highest result for any model. It flags 83.3% of human writing as well, so this says more about TMR than about Fable.
Desklib caught 74.6% of Haiku texts and only 32.5% of Fable texts, a larger gap between two models than the gap between most detectors. Anyone relying on a detector should know which model the text might come from, and with Fable the answer from these detectors is mostly “human”.
Questions people ask
Can AI detectors detect Claude Fable 5.1?
Rarely. The best result from a detector that passes most human writing was Fakespot at 50.0%, and Fakespot also flagged 34.9% of human texts. Desklib, the most careful detector, caught 32.5%.
Is Fable harder to detect than Opus?
Yes. A majority of detectors caught 38.9% of Fable texts and 52.4% of Opus texts.