Voices
Hugging Face's 'State of Open Models: Summer 2026' Puts Local Inference at Trillion-Parameter Scale — Kimi-K3 GGUF at ~2.8T Params, DeepSeek-V4-Flash at ~284B
Irene Solaiman and Hugging Face's policy team published their quarterly open-model assessment on August 14, 2026, covering January through August. Public model repos grew from 2.43M to 2.96M, datasets from 711K to 1M, and Spaces from 1.00M to 1.44M over the period. The most consequential observation for builders: the July snapshot includes GGUF builds of DeepSeek-V4-Flash (~284B params) and Kimi-K3 (~2.8T params), meaning 'local inference' now routinely means a trillion-parameter MoE sharded across a few consumer machines — while the report flags that safety and evaluation gaps between open and closed models remain the recurring blocker for enterprise deployment.
Source
↳ Follow the thread