Seeking Local Model Recommendations Across VRAM Tiers

Teknium · x · 2026-07-10

Teknium is crowdsourcing recommendations for the best local models to run Hermes across different VRAM tiers. Their initial guess suggests Gemma4 12B for 8–16GB VRAM, Qwen3.6 27B/35B for 24–32GB VRAM, and notes they are still looking for optimal options for the 128GB tier.

Original post →

More from Infra

Infra channel →