Leaderboard / Claude Fable 5.1

Claude Fable 5.1 High

Scores

Buzz

• Unverified rumours 😃 Physical AI anxiety and Huawei Ascend pretraining: Sources close to DeepSeek note that OpenAI's GPT 6 Astra hitting 95% accuracy on a robotic control task (versus 40% for Claude Fable 5.1) triggered internal concern that scaling text and code LLMs alone wi…@IamEmily2050 · 28 likes
• Here’s proof, btw. This isn’t even frontier work yet, just some unusual thinking. I’ll test some frontier stuff throughout the rest of the week. I’m willing to go back to Fable 5.1 if needed; it’s still a good colleague. https://t.co/5n29wfZSDb@LarryPanozzo · 3 likes · positive
• Fable 5.2 declared “task complete in 30 trials. It was right once. Every false claim came after real progress, so the work looked finished, which makes it harder to catch. Fable 5.2 made more false claims than any other Claude model: 29, against 11 for Fable 5.1, 3 for Claude …@micro1_ai · 3 likes · negative

Evidence

Coding 63.2 ±2.6

Agents 62.7 ±1.9

Reasoning 64.7 ±1.9

General 68.0 ±2.8

Names matched to this model

claude-fable-5-1-high

Compare Claude Fable 5.1 High with other models