Can I run Qwen-AgentWorld-35B-A3B on an RTX 3060 12GB (12GB)?
Computed with the open FitLLM engine — accurate per-layer KV-cache modeling, not a naive estimate. Updated 2026-10-19.
Conditions: Q4_K_M · KV cache F16 · 32K context · Windows + display reserve (2 GB).
Memory breakdown (Q4_K_M, F16 KV, 32K context)
| Model weights | 19.8 GB |
| KV cache | 0.6 GB |
| Linear-attention state (fixed — does not grow with context) | 0.1 GB |
| Runtime overhead + reserve | 5.4 GB |
| Total used | 25.9 / 12 GB |
| Short by | 13.9 GB |
Context at Q4_K_M: does not fit at any context.
Every quantization on the RTX 3060 12GB
| Weight quant | Weights | Recommended max ctx | Hard limit | Used @32K |
|---|---|---|---|---|
| Q4_K_M | 19.8 GB | won't fit | ❌ 25.9 / 12.0 GB | |
| Q5_K_M | 23.0 GB | won't fit | ❌ 29.6 / 12.0 GB | |
| Q6_K | 26.5 GB | won't fit | ❌ 33.5 / 12.0 GB | |
| Q8_0 | 34.3 GB | won't fit | ❌ 42.2 / 12.0 GB | |
| FP16 | 64.6 GB | won't fit | ❌ 76.2 / 12.0 GB | |
Recommended max = the longest context that still keeps the 5% headroom · hard limit = the longest that fits at all (KV cache F16).
Lower weight quants free memory at some output-quality cost — Q4 is the common sweet spot; below that quality drops faster.
KV cache is F16 here (llama.cpp default). Drop it to Q8/Q4 (-ctk/-ctv) for more context.
Embed this verdict
Live badge for your README or model card — recomputed by the engine at exactly these conditions, never stale:
[](https://fitllm.run/can-i-run/qwen-agentworld-35b-a3b-on-rtx-3060-12gb)
Or from your terminal (exit 0/1 — works as a pre-download guard):
npx fitllm "Qwen-AgentWorld-35B-A3B" --gpu "RTX 3060 12GB" --quant Q4_K_M --ctx 32768 --kv 16Why most VRAM calculators get this wrong
Qwen-AgentWorld-35B-A3B is a hybrid model: only 10 of its 40 layers use full attention — the rest are linear and keep no growing KV cache. Naive calculators count every layer at full context and badly over-estimate.
What fits on the RTX 3060 12GB instead
same GPU Models that fit on the RTX 3060 12GB: Gemma 4 e2b, Gemma 4 e4b, Gemma 4 12b, Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, MiniCPM5-1B, Qwen3-0.6B, Qwen3-1.7B, Llama-3.2-1B-Instruct, Gemma-3-1B-it, Spark-X2.5-4B, MiniCPM5-2B.
same model GPUs that run Qwen-AgentWorld-35B-A3B: RTX 5090 (32GB), RTX 6000 Ada (48GB), RTX PRO 6000 Blackwell (96GB), Radeon PRO W7900 (48GB), 2× RTX 3090 (48GB), 2× RTX 4090 (48GB), 4× RTX 3090 (96GB), A100 40GB (40GB), A100 80GB (80GB), H100 80GB (80GB), H200 141GB (141GB), B200 (180GB), RTX PRO 5000 Blackwell 48GB (48GB), RTX PRO 5000 Blackwell 72GB (72GB), RTX PRO 4500 Blackwell (32GB), DGX Spark 128GB (128GB), 2× DGX Spark 128GB (256GB), 4× DGX Spark 128GB (512GB).
Reproduce it
Qwen-AgentWorld-35B-A3B = 34.7B (3B active, MoE), 40 layers. The RTX 3060 12GB has 12GB / 360GB/s. Same math, open source: fitllm-engine. GGUF bpw from llama.cpp.
All numbers are computed by the open-source fitllm-engine (MIT) from official model config.json values — reproduce or audit them yourself. Estimates; real usage varies with runtime (llama.cpp / MLX / Ollama), driver and display. Found a mismatch? Report it. · FitLLM home