Claude Fable 5 Breaks WeirdML Record
scaling01 · x · 2026-07-12
Claude Fable 5 (max) scored 91.9% on WeirdML, setting a new record. Key points:
- If you combine the historical best scores per task for a theoretical SOTA, the total would be 93.5%, indicating Fable 5 in a single run generally approaches task-specific bests.
- It achieved SOTA on 7 out of 17 tasks.
- Its worst run was only 7% below the corresponding SOTA.
- This run used only 2 trials per task, fewer than the common 5, making the result more impressive but also more likely to dodge a few bad samples.
More from Models
- GPT-6 Astra beats Factorio with enemies in 44 in-game hours at ~$4,500 API cost — liminal_bardo · 2026-09-11
- giffmana: the env being used in training is part of the point — giffmana · 2026-09-11
- awesome-llm-leaderboards: an open-source directory of LLM leaderboards, pricing tables, comparison tools — Last_Establishment_1 · 2026-09-11
- Anthropic claims it works to keep eval environments unidentifiable to models — MaxKannen · 2026-09-11
- Nex N2.5 Pro released on Hugging Face with 407GB of weights — jinnyjuice · 2026-09-11
- RoMa v2 image matching model unveiled in the usual black poster — ducha_aiki · 2026-09-11