llamaperf

Best local LLMs by hardware tier

Rankings only make sense once you fix the hardware. Pick a tier below — the leaderboard re-sorts to the models the community actually runs there, weighted by report count, fastest observed tokens-per-second, and recency.

Looking for a coding model? See the best local LLM for coding by VRAM.

Hardware tier
Model size

Apple Silicon 48–96 GB

Mid-range Apple Silicon. 70B-class fits in Q4 with bandwidth-limited speed. Ranked from 32 reports.

#Model familyReportsFastest t/s
1Qwen3.8Alibaba
T5 · on M3 Max 48GB
1138.0
2DeepSeek V4 FlashDeepSeek
9B distill · 9B · Q4_K_M · on M5 Pro 48GB
844.0
3Qwen3.6Alibaba
27B · 27B · 4-bit · on M5 Max 64GB
663.0
4MuseMeta
Glimmer · 30B · 8-bit · on M4 Pro 48GB
218.0
5Gemma 4Google DeepMind
26B-A4B · 26B-A4B · on M5 Max 64GB
197.0
6Qwen3Alibaba
8B · 8B · Q4 · on M2 Max 96GB
143.0
7LingBot-World-V2
causal-fast · 1.3B · 8-bit · on M4 Pro 48GB
1
8GLM-5.3Zhipu AI
JANG · on M4 Pro 48GB
11.8
9DeepSeek V4.1 Flash
· on M2 Max 64GB
12.0

How we rank

A single global "best models" list doesn't really exist — what runs well on a 5090 is often unrunnable on a 4060, and a 7B that screams on an M3 Max is usually a poor pick on an H100. So we fix the hardware first, then rank the families that actually have community reports on it. The score blends popularity (log-scaled report count), fastest observed tokens-per-second normalized within the bucket, recency (90-day half-life), and a small bias for rows where we know the variant + quant + GPU cleanly. Click into a family for the full breakdown of records.