What Models?

Pick the right model for your GPU — in seconds.

Best local LLMs for 12 GB VRAM (November 2026): what fits, context and speed

With 12 GB of VRAM or unified memory, 42 models run well and 1 more are a tight fit. 64 don't fit.

Speeds assume 444 GB/s, the median memory bandwidth of the 14 GPUs and memory configurations with exactly 12 GB. Yours may be faster or slower: pick your GPU for exact numbers.

Best overall

Highest MMLU-Pro score among models that run well

  1. Phi-4 14B Q4_K_M 16K context · ~44 tok/s · MMLU 84.8
  2. Qwen3.5 9B Q8_0 100K context · ~42 tok/s · MMLU 82.5
  3. Qwen 2.5 14B Q4_K_M 19K context · ~44 tok/s · MMLU 79.9
  4. Qwen3.5 4B Q8_0 252K context · ~81 tok/s · MMLU 79.1
  5. Qwen3 8B Q8_0 28K context · ~46 tok/s · MMLU 79

Best for coding

Highest SWE-bench Verified score among models that run well

  1. K2-Horizon 7B Q8_0 22K context · ~42 tok/s · SWE-bench 72.4
  2. Ornith 1.5 9B Q8_0 93K context · ~41 tok/s · SWE-bench 70.6
  3. MiMo V2.6 Distill Qwen 9B Q8_0 100K context · ~42 tok/s · SWE-bench 61.1
  4. GPT-OSS 20B Q8_0 31K context · ~106 tok/s · SWE-bench 60.7
  5. Mellum2.1 12B A2.5B Thinking Q4_K_M 128K context · ~118 tok/s · SWE-bench 47

GPUs with 12 GB of memory

Graphics cards

Upgrade options

16 GB runs 6 more models well, including Magistral Small 24B, Mistral Small 3.1 24B and Devstral Small 24B.

The cheapest GPUs with more than 12 GB, by launch MSRP.

As an Amazon Associate, I earn from qualifying purchases.

Every model in 12 GB

79 of 214 model configurations can run on 12 GB — sorted by quality

107 models, each counted once by the best way it runs on your hardware.

Largest that runs well GPT-OSS 20B 21B params · Q8_0 · 31K context
Fastest good-quality model Phi-3 Mini 3.8B ~131 tok/s · Q4_K_M · 26K context
? MMLU (Massive Multitask Language Understanding) measures general knowledge across 57 subjects. Higher = more capable. Sorted best-first.

Runs well 42

  • Phi-4 14B 14.7B Excellent
    MMLU 84.8
    Q4_K_M
    Context 16K Speed ~44 tok/s
    Q8_0 Doesn't fit · needs 14.7 GB
  • Qwen3.5 9B 9B Great VisionTool useReasoning
    MMLU 82.5
    Q4_K_M
    Context 216K Speed ~66 tok/s
    Q8_0
    Context 100K Speed ~42 tok/s
  • Qwen 2.5 14B 14.7B Great Tool use
    MMLU 79.9
    Q4_K_M
    Context 19K Speed ~44 tok/s
    Q8_0 Doesn't fit · needs 14.8 GB

Runs well 42

  • Phi-4 14B 14.7B Excellent
    MMLU 84.8
    Q4_K_M
    Context 16K Speed ~44 tok/s
    Q8_0 Doesn't fit · needs 14.7 GB
  • Qwen3.5 9B 9B Great VisionTool useReasoning
    MMLU 82.5
    Q4_K_M
    Context 216K Speed ~66 tok/s
    Q8_0
    Context 100K Speed ~42 tok/s
  • Qwen 2.5 14B 14.7B Great Tool use
    MMLU 79.9
    Q4_K_M
    Context 19K Speed ~44 tok/s
    Q8_0 Doesn't fit · needs 14.8 GB
  • Qwen3.5 4B 4B Great VisionTool useReasoning
    MMLU 79.1
    Q4_K_M
    Context 256K Speed ~119 tok/s
    Q8_0
    Context 252K Speed ~81 tok/s
  • Qwen3 8B 8.2B Great ReasoningTool use
    MMLU 79
    Q4_K_M
    Context 53K Speed ~74 tok/s
    Q8_0
    Context 28K Speed ~46 tok/s
  • Gemma 4 12B 11.96B Great VisionTool useReasoning
    MMLU 77.2
    Q4_K_M
    Context 256K Speed ~55 tok/s
    Q8_0 Tight fit · 13K context
  • Nemotron Nano 12B 12.6B Great ReasoningTool use
    MMLU 76
    Q4_K_M
    Context 128K Speed ~52 tok/s
    Q8_0 Doesn't fit · needs 12.2 GB
  • GPT-OSS 20B 21B Great ReasoningTool use
    MMLU 75
    Q4_K_M
    Context 50K Speed ~108 tok/s

    MoE — speed based on ~1.86 GB active experts plus per-token routing overhead (full 10.83 GB loaded in VRAM)

    Q8_0
    Context 31K Speed ~106 tok/s

    MoE — speed based on ~1.93 GB active experts plus per-token routing overhead (full 11.28 GB loaded in VRAM)

  • Qwen 2.5 7B 7.62B Good Tool use
    MMLU 74.2
    Q4_K_M
    Context 128K Speed ~78 tok/s
    Q8_0
    Context 84K Speed ~49 tok/s
  • Qwen 2.5 VL 7B 7.62B Good VisionTool use
    MMLU 74.2
    Q4_K_M
    Context 128K Speed ~78 tok/s
    Q8_0
    Context 84K Speed ~49 tok/s
  • Gemma 3 12B 11.8B Good Vision
    MMLU 74
    Q4_K_M
    Context 85K Speed ~53 tok/s
    Q8_0 Tight fit · 5K context
  • Qwen 2.5 Coder 14B 14.7B Good
    MMLU 74
    Q4_K_M
    Context 19K Speed ~44 tok/s
    Q8_0 Doesn't fit · needs 14.8 GB
  • Llama 3.2 Vision 11B 10.6B Good VisionTool use
    MMLU 73
    Q4_K_M
    Context 49K Speed ~60 tok/s
    Q8_0 Tight fit · 13K context
  • DeepSeek R1 Distill Qwen 14B 14.7B Good Reasoning
    MMLU 72
    Q4_K_M
    Context 19K Speed ~44 tok/s
    Q8_0 Doesn't fit · needs 14.8 GB
  • Phi-3 Mini 3.8B 3.82B Good
    MMLU 70.9
    Q4_K_M
    Context 26K Speed ~131 tok/s
    Q8_0
    Context 22K Speed ~88 tok/s
  • Phi-4 Mini 3.8B 3.84B Good
    MMLU 70
    Q4_K_M
    Context 79K Speed ~127 tok/s
    Q8_0
    Context 67K Speed ~87 tok/s
  • Gemma 4 E4B 4.5B Good VisionTool useReasoning
    MMLU 69.4
    Q4_K_M
    Context 128K Speed ~69 tok/s
    Q8_0
    Context 128K Speed ~49 tok/s
  • Llama 3.1 8B 8.03B Good Tool use
    MMLU 69.4
    Q4_K_M
    Context 60K Speed ~75 tok/s
    Q8_0
    Context 33K Speed ~47 tok/s
    fp16 Doesn't fit · needs 15.1 GB
  • Qwen 2.5 Coder 7B 7.62B Good
    MMLU 68
    Q4_K_M
    Context 128K Speed ~78 tok/s
    Q8_0
    Context 84K Speed ~49 tok/s
  • Mistral Nemo 12B 12.2B Good Tool use
    MMLU 68
    Q4_K_M
    Context 26K Speed ~52 tok/s
    Q8_0 Doesn't fit · needs 12.3 GB
  • Qwen 2.5 VL 3B 3.09B Fair Vision
    MMLU 65.6
    Q4_K_M
    Context 128K Speed ~151 tok/s
    Q8_0
    Context 128K Speed ~104 tok/s
  • Qwen 2.5 3B 3.09B Fair Tool use
    MMLU 65.6
    Q4_K_M
    Context 32K Speed ~151 tok/s
    Q8_0
    Context 32K Speed ~104 tok/s
  • Ministral 8B 8.02B Fair Tool use
    MMLU 65
    Q4_K_M
    Context 50K Speed ~69 tok/s
    Q8_0
    Context 29K Speed ~47 tok/s
  • SmallThinker 3B 3.09B Fair Reasoning
    MMLU 65
    Q4_K_M
    Context 32K Speed ~143 tok/s
    Q8_0
    Context 32K Speed ~98 tok/s
  • Llama 3.2 3B 3.21B Fair Tool use
    MMLU 63.4
    Q4_K_M
    Context 94K Speed ~147 tok/s
    Q8_0
    Context 82K Speed ~100 tok/s
    fp16
    Context 56K Speed ~60 tok/s
  • Gemma 3 4B 3.88B Fair Vision
    MMLU 63
    Q4_K_M
    Context 128K Speed ~127 tok/s
    Q8_0
    Context 128K Speed ~86 tok/s
  • Gemma 4 E2B 2.3B Fair VisionTool useReasoning
    MMLU 60
    Q4_K_M
    Context 128K Speed ~100 tok/s
    Q8_0
    Context 128K Speed ~74 tok/s
  • DeepSeek R1 Distill Qwen 7B 7.62B Fair Reasoning
    MMLU 58
    Q4_K_M
    Context 128K Speed ~78 tok/s
    Q8_0
    Context 84K Speed ~49 tok/s
  • Qwen3.5 2B 2B Fair VisionTool useReasoning
    MMLU 55.3
    Q4_K_M
    Context 256K Speed ~195 tok/s
    Q8_0
    Context 256K Speed ~148 tok/s
  • StarCoder2 15B 15.6B Basic
    MMLU 51.4
    Q4_K_M
    Context 16K Speed ~41 tok/s
    Q8_0 Doesn't fit · needs 15.9 GB
  • DeepSeek Coder 6.7B 6.74B Basic
    MMLU 50
    Q4_K_M
    Context 16K Speed ~87 tok/s
    Q8_0 Tight fit · 10K context
  • Llama 3.2 1B 1.24B Basic Tool use
    MMLU 49.3
    Q4_K_M
    Context 128K Speed ~246 tok/s
    Q8_0
    Context 128K Speed ~191 tok/s
  • Qwen3.5 0.8B 0.8B Basic VisionTool useReasoning
    MMLU 29.7
    Q4_K_M
    Context 256K Speed ~289 tok/s
    Q8_0
    Context 256K Speed ~244 tok/s
  • Nemotron 3 Nano 4B 3.97B Basic ReasoningTool use
    MMLU
    Q4_K_M
    Context 256K Speed ~112 tok/s
    Q8_0
    Context 256K Speed ~85 tok/s
  • LFM3 1.7B 1.73B Basic Tool useReasoning
    MMLU
    Q4_K_M
    Context 256K Speed ~212 tok/s
    Q8_0
    Context 256K Speed ~156 tok/s
  • Ornith 1.5 9B 9.65B Basic Tool useReasoning
    MMLU
    Q4_K_M
    Context 213K Speed ~66 tok/s
    Q8_0
    Context 93K Speed ~41 tok/s
  • MiMo V2.6 Distill Qwen 9B 9.41B Basic VisionTool useReasoning
    MMLU
    Q4_K_M
    Context 211K Speed ~65 tok/s
    Q8_0
    Context 100K Speed ~42 tok/s
  • LFM2.5 2.6B 2.7B Basic Tool use
    MMLU
    Q4_K_M
    Context 128K Speed ~166 tok/s
    Q8_0
    Context 128K Speed ~115 tok/s
  • Mellum2.1 12B A2.5B Thinking 12.15B Basic Tool useReasoning
    MMLU
    Q4_K_M
    Context 128K Speed ~118 tok/s

    MoE — speed based on ~1.55 GB active experts plus per-token routing overhead (full 7.52 GB loaded in VRAM)

    Q8_0 Doesn't fit · needs 12.1 GB
  • LFM3-VL 2B 1.92B Basic VisionTool useReasoning
    MMLU
    Q4_K_M
    Context 128K Speed ~212 tok/s
    Q8_0
    Context 128K Speed ~156 tok/s
  • Nemotron 3.5 Bolt E4B 8B Basic Tool useReasoning
    MMLU
    Q4_K_M
    Context 64K Speed ~62 tok/s
    Q8_0
    Context 64K Speed ~49 tok/s
  • K2-Horizon 7B 9B Basic Tool useReasoning
    MMLU
    Q4_K_M
    Context 49K Speed ~67 tok/s
    Q8_0
    Context 22K Speed ~42 tok/s

Tight fit 1

  • Gemma 2 9B 9.24B Good
    MMLU 71.3
    Q4_K_M
    Context 8K Speed ~66 tok/s

    Only 8K context — under the 16K needed to be practical

    Q8_0
    Context 8K Speed ~41 tok/s

    Only 8K context — under the 16K needed to be practical

Notify me when new models fit my GPU

An email when a new model fits your 12 GB setup — no spam, unsubscribe anytime.

Doesn't fit 64

  • DeepSeek R1 0528 671B Excellent Reasoning
    MMLU 90.8
    Q4_K_M Needs 381.2 GB — 369.2 GB over
    Q8_0 Needs 664.4 GB — 652.4 GB over
  • Qwen3 Next 80B-A3B 80B Excellent Tool use
    MMLU 90
    Q4_K_M Needs 45.1 GB — 33.1 GB over
    Q8_0 Needs 79.0 GB — 67.0 GB over
    GPUs that can run this

    As an Amazon Associate, I earn from qualifying purchases.

  • MiMo V2.5 Pro 1020B Excellent Reasoning
    MMLU 89.4
    Q4_K_M Needs 586.5 GB — 574.5 GB over
    Q8_0 Needs 1013.0 GB — 1001.0 GB over

Doesn't fit 64