Wire flash
Thom Wolf: 18 of 20 best open models are Mixture-of-Experts
Editorial responsibility
- No named human review is recorded for this page.
- Source reporting is collected, normalized, translated or condensed automatically when needed.
- Automatically published source-backed update
In a post on X, Thom Wolf stated that 18 of the 20 best open models today are Mixture-of-Experts (MoE) architectures, declaring that dense models are "basically over." The post frames MoE as the dominant approach among top-performing open models and implies dense architectures are being displaced. The text is truncated and does not specify which models, benchmarks, or time frame underpin the count, nor does it detail the training know-how referenced in the cut-off sentence. The claim is an attributed opinion/observation from Wolf rather than a formal research release, and no supporting data or methodology is included in the source item.
Source report
18 of the 20 best open models today are Mixture-of-Experts (MoE), according to Thomas Wolf.
Wolf noted that dense models are "basically over," while raising a question about the expertise required to train a good MoE model.
Source
yoheinakajima
Part of this Story
Thom Wolf Says 18 of 20 Best Open Models Are Mixture-of-Experts