How much does it cost to run a frontier-size open model locally?
About $28,000 to $57,000 in Mac hardware as of October 2026, sized with room for context. Three 256 GB Mac Studios ($28,497) run DeepSeek V3.1 or GLM-5.3 at 4-bit. Four ($37,996) run Kimi K2 Thinking. Six ($56,994) run DeepSeek V4 Pro. Holding the same weights on 80 GB NVIDIA H100s takes at least five GPUs for DeepSeek V3.1 and twelve or more for DeepSeek V4 Pro, at roughly $25,000 to $40,000 each.
For reference, Jeff Geerling’s four-Mac-Studio cluster (1.5 TB of memory, about $39,600 in M3 Ultra hardware) ran DeepSeek V3.1 at 32.5 tokens per second and Kimi K2 Thinking at 28.3 tokens per second with exo.
Running costs are small: an M5 Ultra Mac Studio draws at most about 385 W, around $56 a month at $0.20 per kWh even at full load around the clock.
These counts leave headroom. By default macOS lets the GPU use about three quarters of each Mac’s memory (adjustable with the iogpu.wired_limit_mb setting), and each active conversation needs space for its context, so the counts plan on about 160 GB of weights per 256 GB Mac Studio.
| Model | Weights | Mac Studios | Hardware |
|---|---|---|---|
| DeepSeek V3.1 (671B), 4-bit | 378 GB | 3 | $28,497 |
| GLM-5.3 (744B), 4-bit | 418 GB | 3 | $28,497 |
| DeepSeek V4.1 Flash, as released | 510 GB | 4 | $37,996 |
| Kimi K2 Thinking (1T), native 4-bit | 594 GB | 4 | $37,996 |
| DeepSeek V4 Pro (1.6T), as released | 893 GB | 6 | $56,994 |
Sources
- Apple Store: Mac Studio prices (checked Oct 26, 2026)
- Jeff Geerling: 1.5 TB of VRAM on Mac Studio, RDMA over Thunderbolt 5
- Hugging Face: deepseek-ai/DeepSeek-V3.1
- Hugging Face: moonshotai/Kimi-K2-Thinking
- Hugging Face: zai-org/GLM-5.3
- Hugging Face: deepseek-ai/DeepSeek-V4-Pro-0813
- Epoch AI: what a B200 costs
Updated