Claude Fable Refuses Every Task
ProgramBench UPDATED: 6/9/2026 Can language models rebuild programs from scratch? vals.ai VIEW: ALL MODELS TASK TYPE: OVERALL Systems (20) Accuracy Cost/Test Latency Claude Fable 5 Claude Fable 5 refused 200 of 200 tasks (100.00% refusal rate). Click to see the score with fallbacks counted as failures. N/A N/A N/A GPT 5.5 0.50 % ± 0.50 $6.95 1322.96 s GPT 5.4 (high) 0.50 % ± 0.50 $2.27 958.50 s
A dark-mode ProgramBench leaderboard shows Claude Fable 5 refusing all 200 programming tasks, resulting in unavailable scores. The joke is that the model effectively costs money while doing nothing, making the image a reaction to an absurdly unhelpful or overcautious AI.
leaderboard / таблица лидеровcomputer interface / компьютерный интерфейсdark mode screen / экран в тёмном режимеprogramming benchmark / тест программированияtooltip / всплывающая подсказкаtable rows / строки таблицыscore columns / столбцы с оценкамиtext labels / текстовые меткиAI model entries / записи моделей ИИblue badge / синяя плашкаgreen icon / зелёный значок
- @dev_meme / 8110 2026-06-12
- Type
- image
- Size
- 664×410 · 29.2 KiB
- Mood
- саркастичный абсурд
- Language
- en
- Added
