8 GiB GPU

Fits, unrated

• Ministral 3 8B 2512 — Q4_K_M, 7.376 GiB unrated
• Ministral 3 3B 2512 — Q8_0, 5.85 GiB unrated

Does not fit, or not profiled

• cohere/command-a-plus — even Q4_K_M wants 132.618 GiB; pool is 8 GiB
• deepseek/deepseek-v4-flash-vision-exp — even Q4_K_M wants 184.902 GiB; pool is 8 GiB
• z-ai/glm-5.3 — even Q4_K_M wants 455.624 GiB; pool is 8 GiB
• qwen/qwen3.8-27b — even Q4_K_M wants 18 GiB; pool is 8 GiB
• qwen/qwen3.8-2.4t-a95b — even Q4_K_M wants 1450.719 GiB; pool is 8 GiB
• meta/muse-glimmer-30b — even Q4_K_M wants 19.277 GiB; pool is 8 GiB
• moonshotai/kimi-k3 — even Q4_K_M wants 1691.5 GiB; pool is 8 GiB
• z-ai/glm-5.2 — even Q4_K_M wants 455.624 GiB; pool is 8 GiB
• moonshotai/kimi-k2.7-code — even Q4_K_M wants 604.75 GiB; pool is 8 GiB
• nvidia/nemotron-3-ultra-550b-a55b — even Q4_K_M wants 333.063 GiB; pool is 8 GiB
• minimax/minimax-m3 — even Q4_K_M wants 259.405 GiB; pool is 8 GiB
• deepseek/deepseek-v4-pro — even Q4_K_M wants 967 GiB; pool is 8 GiB
• deepseek/deepseek-v4-flash — even Q4_K_M wants 172.465 GiB; pool is 8 GiB
• google/gemma-4-31b-it — even Q4_K_M wants 20.6 GiB; pool is 8 GiB
• mistralai/mistral-small-2603 — even Q4_K_M wants 73.2 GiB; pool is 8 GiB
• mistralai/ministral-14b-2512 — even Q4_K_M wants 10.642 GiB; pool is 8 GiB
• mistralai/mistral-large-2512 — even Q4_K_M wants 408.531 GiB; pool is 8 GiB
• deepseek/deepseek-v3.2 — even Q4_K_M wants 406.116 GiB; pool is 8 GiB
• openai/gpt-oss-120b — even Q4_K_M wants 72.201 GiB; pool is 8 GiB
• meta-llama/llama-4-maverick — even Q4_K_M wants 242.5 GiB; pool is 8 GiB
• meta-llama/llama-4-scout — even Q4_K_M wants 66.809 GiB; pool is 8 GiB
• cohere/command-a — even Q4_K_M wants 68.016 GiB; pool is 8 GiB
• microsoft/phi-4 — even Q4_K_M wants 11.015 GiB; pool is 8 GiB

GPU

How the number is made

Weights: a measured GGUF size when we have one, otherwise total parameters × bits per weight ÷ 8. Q4_K_M is treated as 4.83 bits/param. MoE memory uses total parameters, not active parameters.

KV cache, when layer and head geometry is sourced: 2 × kvLayers × kvHeads × headDim × context × 2 bytes (FP16, batch 1). Hybrid models use attention-layer count, not every layer. If geometry is missing, cache is omitted and that is stated.

1 GiB is added for CUDA/runtime. Apple unified memory uses RAM as the pool. CPU-only is a memory fit, not a speed claim. DDR generation is ignored for fit — it changes bandwidth, not whether the weights sit in memory.

RAM type (DDR4/DDR5/LPDDR) is not a field. It does not change whether a model fits.

Self-host