Best local LLMs for 12 GB of VRAM or unified memory

Measured speeds from builds where the model ran on more than 8 and up to 12 GB of GPU or unified memory. Each number links to its build and source.

8 GB · 12 GB · 16 GB · 24 GB · 32 GB · 48 GB · 64 GB · 96 GB · 128 GB or more

GPUs in this tier: RTX 3060 12GB

Fastest measured result per model

Excludes results the source marks pending and results where part of the model sat in system RAM.

All results in this tier

Hide speculative-decoding results ⚡ (MTP, DFlash, DSpark and similar draft-and-verify methods give much higher tok/s on predictable text and are not comparable 1:1 with plain decode)

What fits by memory math

Q4_K_M weights (4.8944 bits per weight, per the llama.cpp quantize table) that fit in 12 GB with ~5% headroom, before KV cache: Mellum 2.1 12B-A2.5B ≈7 GB, Gemma 4 12B ≈7 GB, Qwen3.5 9B ≈5 GB, Llama 3.1 8B legacy ≈5 GB.

FAQ

Which local LLMs have been measured on 12 GB of VRAM?

DeepSeek R1 Distill 14B, Llama 3.2 3B, gpt-oss 20B, Gemma 4 12B, LFM2.5 24B-A2B, Qwen3.8 27B, Qwen3.8 Flash-Next, Mellum 2.1 12B-A2.5B (16 results).

What fits in 12 GB of VRAM?

By memory math at Q4_K_M (4.8944 bits per weight, llama.cpp), weights only, before KV cache and context: Mellum 2.1 12B-A2.5B (≈7 GB), Gemma 4 12B (≈7 GB), Qwen3.5 9B (≈5 GB), Llama 3.1 8B (≈5 GB). Leave a few GB for context.

Are speculative-decoding results comparable?

Not 1:1. MTP, DFlash and similar methods draft tokens that the main model verifies, which can double tok/s on predictable text. This page lists them in a separate column; tick the box to hide them in the full table.