Can 16-32GB Local Models Match Closed-Source SOTA by End of 2025?
AnimalPuzzleheaded71 · reddit · 2026-11-22
A Reddit user asks whether "true local" models — not the 500B-parameter open-weight models no ordinary person can run, but those that realistically run at 10-20 tokens/s on 16-32GB of regular RAM and 8-16GB of VRAM — have caught up with state-of-the-art closed models by the end of 2025. The poster notes this range covers most consumer PCs and wants to gauge the real gap between this tier of local models and top closed-source models.
More from Infra
- Speed-to-power dominates: on-site solar-plus-storage cheapest for AI data centers in 85% of US locations — mattbeane · 2026-11-22
- LexiEngine, an AMD-focused Rust inference engine, adds Inkling support — W61k3r · 2026-11-22
- 9B model hits 59 tok/s on a 16GB GPU, but tool calls lag at 58% BFCL — AIFlow_ML · 2026-11-22
- Running 17.6GB Inkling 2 Mini on a 12GB phone at 3 tok/s — dai_app · 2026-11-22
- NVIDIA Paper Caches LLM Reasoning as CPU Code, 16-291x Cheaper — rohanpaul_ai · 2026-11-22
- Async RL's max_staleness=3 Doesn't Really Cap Staleness: 10-Framework Analysis — 青稞AI · 2026-11-22