Claude Fable 5が全業界ベンチマークを制覇 ― しかしDeepSeekより100倍優れているのか?
AnthropicのClaude Fable 5はArtificial Analysisの6つの新業界ベンチマークで首位を獲得したが、タスクあたり$3.48のコストがかかる一方、DeepSeek V4は$0.04未満。この価格差は、わずかな品質向上のために莫大な費用を正当化できるのか、企業に疑問を投げかけている。
claude fable 5 leads all eight industry benchmarks released by artificial analysis, including specific tests for finance, law, and engineering, based on us o*net occupational standards. while claude opus 4.8 and gpt-5.5 follow closely in performance, the open-source glm-5.2 leads other open-weight models. the findings expose a steep cost premium: claude fable 5 costs roughly $3.48 for specific tasks, whereas deepseek v4 flash handles tasks for less than $0.04. this data validates a trend where cheaper models are increasingly viable, pressuring enterprises to adopt cascading architectures that use premium models only when necessary to optimize spending.