What can NVIDIA GeForce RTX 3060 12GB run for agents?

Build: NVIDIA GeForce RTX 3060 12GB + — + 32 GB RAM (windows)

Memory: 12 GB VRAM + 32 GB system RAM
Runner: llama.cpp / Ollama (CUDA)
AnyChatCodingAgentsReasoningVisionLong contextCreative

Runs comfortably
8 models

Ranked by fit for agents use case + predicted speed. Click a row for VRAM breakdown.

#1Qwen 2.5 7B Instruct
7B
qwen
Commercial OK
Quant: Q5_K_MContext: 8,192VRAM: 7.9 GBHeadroom: 4.1 GBTTFT: noticeable
ollama run qwen2.5:7b
49
tok/s
Estimated
Weights
5.40 GB
KV cache
0.47 GB
Activations
0.28 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~1411 ms (noticeable)
Model details →
#2Trendyol LLM 7B Chat v0.1
7B
llama
Quant: Q4_K_MContext: 4,096VRAM: 7.6 GBHeadroom: 4.4 GBTTFT: noticeable
55
tok/s
Estimated
Weights
3.90 GB
KV cache
1.75 GB
Activations
0.20 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~1411 ms (noticeable)
Model details →
#3Mistral 7B Instruct v0.1
7B
mistral
Commercial OK
Quant: Q4_K_MContext: 4,096VRAM: 7.6 GBHeadroom: 4.4 GBTTFT: noticeable
55
tok/s
Estimated
Weights
3.90 GB
KV cache
1.75 GB
Activations
0.20 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~1411 ms (noticeable)
Model details →
#4Trendyol LLM 7B Base v0.1
7B
llama
Quant: Q4_K_MContext: 4,096VRAM: 7.6 GBHeadroom: 4.4 GBTTFT: noticeable
55
tok/s
Estimated
Weights
3.90 GB
KV cache
1.75 GB
Activations
0.20 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~1411 ms (noticeable)
Model details →
#5Swallow 7B
7B
llama
Quant: Q4_K_MContext: 4,096VRAM: 7.6 GBHeadroom: 4.4 GBTTFT: noticeable
55
tok/s
Estimated
Weights
3.90 GB
KV cache
1.75 GB
Activations
0.20 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~1411 ms (noticeable)
Model details →
#6Bielik 7B Instruct v0.1 GGUF
7B
mistral
Quant: Q4_K_MContext: 4,096VRAM: 7.6 GBHeadroom: 4.4 GBTTFT: noticeable
55
tok/s
Estimated
Weights
3.90 GB
KV cache
1.75 GB
Activations
0.20 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~1411 ms (noticeable)
Model details →
#7Bielik 7B v0.1
7B
mistral
Commercial OK
Quant: Q4_K_MContext: 4,096VRAM: 7.6 GBHeadroom: 4.4 GBTTFT: noticeable
55
tok/s
Estimated
Weights
3.90 GB
KV cache
1.75 GB
Activations
0.20 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~1411 ms (noticeable)
Model details →
#8OpenThaiGPT 7B 1.0.0 Chat
7B
llama
Commercial OK
Quant: Q4_K_MContext: 4,096VRAM: 7.6 GBHeadroom: 4.4 GBTTFT: noticeable
55
tok/s
Estimated
Weights
3.90 GB
KV cache
1.75 GB
Activations
0.20 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~1411 ms (noticeable)
Model details →

Runs with tradeoffs
161 models

Tight VRAM, partial CPU offload, or context-limited.

Hermes 3 Llama 3.1 8B
8B
hermes
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 11.0 GBHeadroom: 1.0 GBTTFT: noticeable
  • • Tight VRAM fit — only 1.0 GB headroom left for context growth
ollama run hermes3:8b
48
tok/s
Estimated
Weights
4.90 GB
KV cache
4.00 GB
Activations
0.25 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~1613 ms (noticeable)
Model details →
Dolphin 3.0 Mistral 24B
24B
dolphin
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 28.5 GBHeadroom: 2.7 GBTTFT: slow
  • • Partial CPU offload: ~58% of layers run on CPU
ollama run dolphin-mistral:24b
16
tok/s
Estimated
Weights
14.00 GB
KV cache
12.00 GB
Activations
0.71 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~4838 ms (slow)
Model details →
Qwen 2.5 14B Instruct
14B
qwen
Commercial OK
Quant: Q8_0Context: 8,192VRAM: 25.3 GBHeadroom: 5.9 GBTTFT: slow
  • • Partial CPU offload: ~53% of layers run on CPU
ollama run qwen2.5:14b
16
tok/s
Estimated
Weights
15.70 GB
KV cache
7.00 GB
Activations
0.79 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~2822 ms (slow)
Model details →
Qwen 3 30B-A3B
30B
qwen
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 24.5 GBHeadroom: 6.7 GBTTFT: slow
  • • Partial CPU offload: ~51% of layers run on CPU
ollama run qwen3:30b
13
tok/s
Estimated
Weights
18.00 GB
KV cache
3.75 GB
Activations
0.90 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~6047 ms (slow)
Model details →
Qwen3 Coder 30B-A3B
30B
qwen
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 25.5 GBHeadroom: 5.7 GBTTFT: slow
  • • Partial CPU offload: ~53% of layers run on CPU
ollama run qwen3-coder:30b
13
tok/s
Estimated
Weights
19.00 GB
KV cache
3.75 GB
Activations
0.95 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~6047 ms (slow)
Model details →
Qwen 2.5 Coder 32B Instruct
32B
qwen
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 23.9 GBHeadroom: 7.3 GBTTFT: slow
  • • Partial CPU offload: ~50% of layers run on CPU
ollama run qwen2.5-coder:32b
12
tok/s
Estimated
Weights
19.00 GB
KV cache
2.15 GB
Activations
0.96 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~6450 ms (slow)
Model details →
Qwen 3 32B
32B
qwen
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 25.8 GBHeadroom: 5.4 GBTTFT: slow
  • • Partial CPU offload: ~53% of layers run on CPU
ollama run qwen3:32b
12
tok/s
Estimated
Weights
19.00 GB
KV cache
4.00 GB
Activations
0.95 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~6450 ms (slow)
Model details →
Qwen 2.5 32B Instruct
32B
qwen
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 25.8 GBHeadroom: 5.4 GBTTFT: slow
  • • Partial CPU offload: ~53% of layers run on CPU
ollama run qwen2.5:32b
12
tok/s
Estimated
Weights
19.00 GB
KV cache
4.00 GB
Activations
0.95 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~6450 ms (slow)
Model details →

What if you upgraded?

Hypothetical scenarios. We re-ran the compatibility engine for each.

+32 GB system RAM

~$80–150

Doubles your CPU-offload working set. Helps when models don't quite fit in VRAM.

Unlocks: 59 new comfortable, 174 new tradeoff

  • • Qwen 3 0.6B
  • • Llama 3.2 3B Instruct
  • • Qwen 3 1.7B
  • • Gemma 3 270M
Shop this upgrade↗

Upgrade to NVIDIA GeForce RTX 4070 Ti Super

~$829

16 GB VRAM (vs your 12 GB) plus a bandwidth jump from ~360 GB/s to ~672 GB/s.

Unlocks: 134 new comfortable

  • • Qwen 3 0.6B
  • • Llama 3.1 8B Instruct
  • • Qwen 3 8B
  • • Llama 3.2 3B Instruct
Shop this upgrade↗

Add a second NVIDIA GeForce RTX 3060 12GB

~$249

Tensor parallelism splits the model across both cards, effectively doubling VRAM. Bandwidth doesn't double — runs ~1.5× the single-card speed in practice.

Unlocks: 176 new comfortable

  • • Qwen 3 0.6B
  • • Llama 3.1 8B Instruct
  • • Qwen 3 8B
  • • Gemma 4 12B
Shop this upgrade↗

Some links above are affiliate links. We may earn a commission at no extra cost to you. How we make money.

Won't run
top 5 popular models

Need more memory than you have. Shown for orientation.

DeepSeek V4 Pro (1.6T MoE)
1600B
deepseek
Commercial OK

Even with CPU offload, needs more memory than your VRAM (12 GB) + 60% of system RAM (19 GB) combined.

—
Qwen 3.5 235B-A17B (MoE)
397B
qwen
Commercial OK

Even with CPU offload, needs more memory than your VRAM (12 GB) + 60% of system RAM (19 GB) combined.

—
Qwen 3 235B-A22B
235B
qwen
Commercial OK

Even with CPU offload, needs more memory than your VRAM (12 GB) + 60% of system RAM (19 GB) combined.

—
DeepSeek R1 (671B reasoning)
671B
deepseek
Commercial OK

Even with CPU offload, needs more memory than your VRAM (12 GB) + 60% of system RAM (19 GB) combined.

—
Llama 4 Scout
109B
llama
Commercial OK

Even with CPU offload, needs more memory than your VRAM (12 GB) + 60% of system RAM (19 GB) combined.

—

How to read these numbers

Measured here
Measured here - RunLocalAI ran this exact combo on owner hardware with public evidence.

Source-backed
Source-backed / community - a reproduced public source supports the speed, but it is not labeled as owner-measured.

Extrapolated
Extrapolated - predicted from a measured benchmark on similar-bandwidth hardware.

Estimated
Estimated - formula based on VRAM bandwidth and model architecture; not a benchmark row.

RunLocalAI Will-It-Run Framework →

Want a specific benchmark we don't have? Email Contact support and we'll prioritize it.