Magic claims >10x more efficient pretraining, matches DeepSeek V4 Pro with 50x fewer FLOPs

Dr_Singularity · x · 2026-09-09

Startup Magic says its new pretraining recipe is over 10x more compute-efficient than leading open-weight models.

Related event: Magic Claims 50x Pretraining Efficiency, Matching DeepSeek V4 Pro for ~$500K(9 posts)→

Original post →

More from Infra

Infra channel →