Free Tool

Can I Run This Model?

Select your hardware and quantization level to see which open-source models you can run locally, how much RAM they need, and estimated generation speed.

Available: 32 GBBandwidth: 400 GB/s
Model
Size
Status
Phi-4 Mini (3.8B)
MIT
3.1 GB
Gemma 3 (4B)
Gemma
3.3 GB
Qwen3.6 (7B)
Apache 2.0
4.9 GB
Llama 4 Scout (8B)
Llama 4
5.5 GB
Gemma 3 (12B)
Gemma
7.8 GB
Qwen3.6 (14B)
Apache 2.0
8.9 GB
Phi-4 (14B)
MIT
8.9 GB
Gemma 3 (27B)
Gemma
16.2 GB
Qwen3.6 (32B)
Apache 2.0
19.0 GB
DeepSeek V4-Flash (284B MoE, 13B active)
MIT
162.8 GB
MiniMax M3 (428B MoE, 23B active)
MiniMax Community
243.8 GB
Tencent Hy3 (295B MoE, 21B active)
Modified MIT
168.9 GB
GLM-5.2 (744B MoE, 40B active)
MIT
421.5 GB
Llama 3.3 (70B)
Llama 3.3
41.4 GB
Qwen3.6 (72B)
Apache 2.0
42.5 GB
DeepSeek V4-Pro (1.6T MoE, 49B active)
MIT
903.0 GB
Kimi K2.7-Code (1T MoE, 32B active)
Modified MIT
565.5 GB
Qwen3.6 (235B MoE, 22B active)
Apache 2.0
135.2 GB
DeepSeek V3 (671B MoE, 37B active)
MIT
380.4 GB
Kimi K3 (2.8T MoE, 16B active)
Modified MIT
1578.0 GB
Runs comfortably (under 75% RAM) Fits but tight (75-100% RAM, may need swap) Exceeds available memory

How this works

How are model sizes calculated?

Model size = (total parameters × bits per parameter) ÷ 8, plus overhead for KV cache and metadata. For example, GLM-5.2 (744B MoE) at 4-bit needs roughly (744 × 4.5) ÷ 8 ≈ 415 GB of RAM — all experts must be loaded. MoE models use total parameters for memory sizing but only active parameters for speed estimation.

How accurate are the speed estimates?

For dense models, tokens/sec ≈ memory_bandwidth ÷ (2 × model_size). For MoE models, we adjust the effective size based on active parameters plus shared layers (~30%). This is a simplified model — actual speeds vary by ±30% depending on the inference engine (llama.cpp, vLLM, MLX), batch size, context length, and expert routing efficiency.

Why does memory bandwidth matter?

LLM inference is memory-bound, not compute-bound. Generating each token requires loading the entire model from memory. This is why Apple Silicon (with 400+ GB/s unified memory) can outperform PCs with the same RAM but slower memory, and why GPUs with HBM (1000+ GB/s) are dramatically faster.

Can I run models that say 'too big'?

Sometimes, with CPU offloading. llama.cpp supports partial GPU offloading where some layers run on GPU and the rest on CPU. This works but is much slower (5-15 tok/s instead of 30-80). It's usable for batch processing but painful for interactive chat.

What about MoE (Mixture of Experts) models?

MoE models like DeepSeek V4-Pro (1.6T total, 49B active) or GLM-5.2 (744B total, 40B active) need memory for ALL parameters — every expert must be in RAM. But during inference, only the active experts are loaded per token, making them much faster than a dense model of the same total size. A 744B MoE at 4-bit needs ~415 GB of RAM but generates tokens as fast as a ~40B dense model. This is why MoE has become the dominant architecture for frontier open models.

Don't want to deal with hardware requirements? CopperRiver handles model hosting for you →