Claude Fable 5 ban showed no productivity boost

arxiv.org1 min read

Researchers investigated whether higher-scoring coding models actually improve real-world developer productivity by treating a 19-day U.S. export control ban on Claude Fable 5 as a natural experiment. By tracking GitHub pull requests of former Fable 5 users who switched to lower-benchmark models, along with upgrades across multiple models, they found that access to higher-scoring models did not significantly increase 7-day merge rates or the number of pull requests merged per week. Open article →