Claude Fable · 发布历史
Anthropic Claude Fable 系列已收录 2 次发布和 49 个测试版本。
Claude Fable 是面向高难推理与长时程 agent 工作的通用分支;Anthropic 说明它与 Mythos 使用同一底层模型,但采用面向一般可用场景的 safeguards。
全部版本
- Claude Fable 5.236 项测试
- Claude Fable 5.137 项测试
已收录测试 · 49
- FrontierCode · 2.0Claude Fable 5.2、Claude Fable 5.1
- FrontierSWE · v2Claude Fable 5.2、Claude Fable 5.1
- AutomationBench · 1.0.6; 47 tools; chart-reported score, scoring rule unspecifiedClaude Fable 5.2
- Terminal-Bench-Science · 0.1Claude Fable 5.2、Claude Fable 5.1
- WANDR · relaxed score;benchmark version 未注明Claude Fable 5.2、Claude Fable 5.1
- OSWorld · 2.1, 82-task offline no-internet subset, partial scoreClaude Fable 5.2、Claude Fable 5.1
- RoboDojo Sim · 版本/任务配置未注明Claude Fable 5.2、Claude Fable 5.1
- AA-Briefcase · v1.1Claude Fable 5.2、Claude Fable 5.1
- Remote Labor Index · 版本未注明Claude Fable 5.2、Claude Fable 5.1
- Humanity's Last Exam · with tools;题集版本未注明Claude Fable 5.2、Claude Fable 5.1
- Chartography · no tools;task/version 未注明Claude Fable 5.2、Claude Fable 5.1
- SWE-bench Pro · ProClaude Fable 5.2、Claude Fable 5.1
- BenchCAD · 2.1 core scoresClaude Fable 5.2、Claude Fable 5.1
- OfficeQA · Full setClaude Fable 5.2、Claude Fable 5.1
- OfficeQA · Pro 133-question subsetClaude Fable 5.2、Claude Fable 5.1
- SWE-bench Multilingual · 300 problems, 9 programming languagesClaude Fable 5.2、Claude Fable 5.1
- SWE-bench Multimodal · MultimodalClaude Fable 5.2、Claude Fable 5.1
- DeepSWE · v1.1Claude Fable 5.2、Claude Fable 5.1
- Terminal-Bench · 4.0Claude Fable 5.2、Claude Fable 5.1
- CursorBench · 4.0Claude Fable 5.2
- STEM Professor Bench · 64-problem internal research subset; score averaged over up to nine attempts per problemClaude Fable 5.2
- ProgramBench · 166 个保留的 golden tasks;架构与时限按协议区分Claude Fable 5.2、Claude Fable 5.1
- Humanity's Last Exam · no tools;题集版本未注明Claude Fable 5.2、Claude Fable 5.1
- DRACO · 100-task Perplexity deep-research benchmarkClaude Fable 5.2、Claude Fable 5.1
- WANDR · 500-task wide research benchmarkClaude Fable 5.2
- Chartography · 100 tasksClaude Fable 5.2、Claude Fable 5.1
- Chartography · Fable 5.2 contamination check; 100 original versus redrawn charts with new data/answers; no tools; five runsClaude Fable 5.2
- Harvey LAB-AA · Artificial Analysis v1.1; 120-task held-out set; hallucination-gated and ungated all-pass ratesClaude Fable 5.2
- GDPval-AA · v2.1Claude Fable 5.2
- Real-World Finance · v2; 123 real-world financial tasks; one attempt per task in the reported Fable 5.2 runClaude Fable 5.2
- ARC-AGI-3 · 版本/子集未注明Claude Fable 5.2
- HealthBench · 48,000+ rubric items across 5,000 multi-turn conversationsClaude Fable 5.2、Claude Fable 5.1
- HealthBench Professional · 525 physician-authored conversationsClaude Fable 5.2、Claude Fable 5.1
- PhysicianBench · 100 physician tasks in an electronic health record; pass rate across attemptsClaude Fable 5.2
- Global MMLU · 42 languages including English · overall averageClaude Fable 5.2
- Multi-task Indic Language Understanding Benchmark · 10 Indic languages plus English · overall averageClaude Fable 5.2
- GDPval-AA · v2Claude Fable 5.1
- OSWorld · 2.0,August 2026 task release(与此前 task files 不可直接比较)Claude Fable 5.1
- AutomationBench · 版本、任务集未在 Fable 发布表注明Claude Fable 5.1
- CursorBench · 3.2.0Claude Fable 5.1
- FrontierCode · v1.1 Extended setClaude Fable 5.1
- FrontierCode · v1.1 Main setClaude Fable 5.1
- BenchCAD Vision2Code · 1,000-file random subset of 17,900 filesClaude Fable 5.1
- CritPT-Corrected · 71 theoretical-physics problems; 31 statements corrected for underspecification, ambiguity, or typos; Pass@1 averaged over 16 attempts per problemClaude Fable 5.1
- GDP.pdf · 100 professional PDFs and promptsClaude Fable 5.1
- Legal Agent Benchmark · Public full setClaude Fable 5.1
- Toolathlon-Verified · June 2026 finalized releaseClaude Fable 5.1
- ARC-AGI-1 · ARC Prize Foundation private datasetClaude Fable 5.1
- ARC-AGI-2 · ARC Prize Foundation private datasetClaude Fable 5.1