哈佛论文:失去 Claude Fable 5 的 19 天里,开发者产出几乎不变
哈佛用 297 名用户被断供 19 天这个天然实验,测出顶级模型对真实代码合并率的贡献只有 1.4 个百分点。选模型前先看看这个,可能帮你省下不少预算。
一篇哈佛论文研究了 2026 年 6 月一项美国出口管制指令导致 Claude Fable 5 停用 19 天的影响。研究追踪了 297 名 Fable 5 用户的 GitHub pull requests,并与 8,908 名留在 Opus 4.8 上的开发者对比。
结果显示,使用 Fable 5 仅将 7 天合并率提高了 1.4 个百分点,处于统计噪声范围内;在被停用期间,这些用户的 PR 在更弱的 Opus 4.8 上合并频率大致相当。研究还检查了另外 5 次模型升级,合并率同样没有上升,原因是 94.5% 的 pull requests 本来就会被合并。
这项研究对'前沿模型能否实质提升开发者生产力'提出了量化质疑:在代码审查和迭代流程已高度成熟的现实项目中,模型能力的边际差异很难体现在最终交付上。对采购决策者和开发者都有参考价值。
New Harvard paper finds that when Claude Fable 5 was suspended for 19 days, its users' pull requests got merged about as often on the weaker Opus 4.8.
A U.S. export control directive cut off Claude Fable 5 for 19 days in June 2026, and most users switched to Opus 4.8. The study tracked GitHub pull requests from 297 Fable 5 users against 8,908 developers who stayed on Opus 4.8.
Access to Fable 5 raised the 7-day merge rate by only 1.4 points, a change within noise. Across 5 other model upgrades, merge rates did not rise either, though 94.5% of pull requests were already merged beforehand.
This report is categorized under AITOP's "Papers" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Claude Fable 5, Opus 4.8, Harvard, GitHub.
Source attribution: curated from rohanpaul_ai (@rohanpaul_ai).