How much VRAM do you need for a 1T+ model?
Short answer: about 716.6 GB at Q4_K_M, the usual quant for local use. Below: every quant, longer contexts, the cheapest hardware that fits, and every 1T+ open model we track.
- Q4_K_M
- 716.6 GB 622.3 GB weights
- Q8_0
- 1255.1 GB 1090.6 GB weights
- FP16
- 2361.7 GB 2052.8 GB weights
No single consumer GPU we track fits it at Q4.
Cheapest hardware for each quant
| Setting | Memory needed | Cheapest GPU | Cheapest Mac / mini PC |
|---|---|---|---|
| Q4_K_M, ~4k context | 716.6 GB | None we track | None we track |
| Q8_0, ~4k context | 1255.1 GB | None we track | None we track |
| FP16, ~4k context | 2361.7 GB | None we track | None we track |
No single consumer GPU we track holds a 1T+ model at Q4. Your options: two or more GPUs (see the ready-made builds or the planner), a large unified-memory machine (see the best Mac for local LLMs), or a smaller quant with lower quality.
Hardware that runs it
Bar: Q4 memory needed vs. usable memory. Speed is estimated decode at Q4 (how we estimate); MoE shows an upper bound only.FitsTight
None of the devices we track run it at Q4.
1T+ models we track
| Model | Q4_K_M | Q8_0 | FP16 |
|---|---|---|---|
| Muse-Spark-Open-816B-A42B822.0B params · meta-models · MoE | 574.1 GB | 1005.4 GB | 1891.7 GB |
| Intern-S1-Pro917.0B params · internlm · MoE | 640.4 GB | 1121.5 GB | 2110.2 GB |
| Inkling952.4B params · thinkingmachines · MoE | 665.0 GB | 1164.7 GB | 2191.5 GB |
| Ling-1T999.7B params · inclusionAI · MoE | 698.0 GB | 1222.5 GB | 2300.3 GB |
| Ring-1T999.7B params · inclusionAI · MoE | 698.0 GB | 1222.5 GB | 2300.3 GB |
| Ring-2.5-1T1012.2B params · inclusionAI · MoE | 706.7 GB | 1237.8 GB | 2329.1 GB |
| Ling-2.5-1T1012.2B params · inclusionAI · MoE | 706.7 GB | 1237.8 GB | 2329.1 GB |
| MiMo-V2.5-Pro1023.2B params · XiaomiMiMo · MoE | 714.4 GB | 1251.2 GB | 2354.3 GB |
| MiMo-V2.6-Pro-RL1024.2B params · XiaomiMiMo · MoE | 715.1 GB | 1252.5 GB | 2356.7 GB |
| MiMo-V2.6-Pro-MOPD1024.2B params · XiaomiMiMo · MoE | 715.1 GB | 1252.5 GB | 2356.7 GB |
| Kimi-K2-Instruct1026.4B params · moonshotai · MoE | 716.6 GB | 1255.1 GB | 2361.7 GB |
| Kimi-K2-Thinking1026.4B params · moonshotai · MoE | 716.6 GB | 1255.1 GB | 2361.7 GB |
| Kimi-K2-Instruct-09051026.4B params · moonshotai · MoE | 716.6 GB | 1255.1 GB | 2361.7 GB |
| Kimi-K2.51026.9B params · moonshotai · MoE | 716.9 GB | 1255.7 GB | 2362.8 GB |
| Kimi-K2.61026.9B params · moonshotai · MoE | 716.9 GB | 1255.7 GB | 2362.8 GB |
| Kimi-K2.7-Code1026.9B params · moonshotai · MoE | 716.9 GB | 1255.7 GB | 2362.8 GB |
| DeepSeek-V4-Pro1598.8B params · deepseek-ai · MoE | 1115.7 GB | 1954.6 GB | 3678.3 GB |
| DeepSeek-V4-Pro-08131650.5B params · deepseek-ai · MoE | 1151.7 GB | 2017.7 GB | 3797.1 GB |
| LongCat-2.01775.6B params · meituan-longcat | 1238.9 GB | 2170.5 GB | 4084.8 GB |
| LongCat-2.51779.8B params · meituan-longcat · MoE | 1241.8 GB | 2175.6 GB | 4094.4 GB |
| Qwen3.8-2.4T-A95B2446.2B params · Qwen · MoE | 1706.4 GB | 2989.9 GB | 5627.2 GB |
| Kimi-K32779.9B params · moonshotai · MoE | 1939.1 GB | 3397.7 GB | 6394.8 GB |
| Kimi-K3.1-Pro2780.1B params · moonshotai · MoE | 1939.3 GB | 3397.9 GB | 6395.3 GB |
Quick answers
- How much VRAM do I need for a 1T+ model?
- About 716.6 GB at Q4_K_M, 1255.1 GB at Q8_0 and 2361.7 GB at FP16, at a ~4k context. That is for Kimi-K2-Instruct (1026.4B parameters): weights plus about 15% for the KV cache and 1 GB for the runtime.
- What is the cheapest GPU for a 1T+ model?
- None of the single consumer GPUs we track hold it at Q4. Use several GPUs, a large unified-memory machine, or CPU offload (slow).
- Can a Mac run a 1T+ model?
- None of the Macs or mini PCs we track fit it at Q4.
- Does a MoE model of the same size need less VRAM?
- No. A mixture-of-experts model still has to hold all of its weights in memory, so it needs about the same VRAM as a dense model with the same total parameters. It runs faster, because only a few experts are read per token.
Other sizes: 1B–2B · 2B–3B · 4B–6B · 7B–8B · 12B–14B · 16B–21B · 24B–27B · 30B–32B · 40B–50B · 70B · 120B · 200B–300B · 300B–480B · 500B–800B · All sizes · Ollama VRAM by tag · GGUF sizes by quant