Open source closes the gap with closed labs: Quality gap now just months

togethercompute · x · 2026-08-15

Together AI cites Tri Dao's remarks at ICML, stating that open-source models are rapidly closing the gap with closed-source labs, lagging by only a few months in quality. Furthermore, advancements in the serving stack (kernels, inference engines, speculative decoding) are delivering compounding improvements.

Original post →

More from Infra

Infra channel →