Cheapest box to run gpt-oss-20b at Q3_K_M
OpenAI’s gpt-oss-20b is 21B parameters, 3.6B of them read per token. The number below is computed from that and each vendor’s published specs, not asserted. Switch the quant or the context and it recomputes in place.
gpt-oss-20b at Q3_K_M needs 12GB.
Cheapest Config that fits: Mac mini, M6, 24GB. Fits, 9GB headroom. Est. 48.3 tok/s.
Best offer today: $1,299 at Apple — in stock, checked 10 hours ago.
Collecting price history — 2 of 90 days. No timing verdict yet.
Offers — Mac mini, M6, 24GB, 512GB
- Applelist pricein stockchecked 10 hours ago$1,299
- Mac mini, M6, 16GB$1,099Tight1GB headroomest. 48.3 tok/s
- Mac mini, M6, 24GB$1,299Fits9GB headroomest. 48.3 tok/s
- Mac mini, M6, 32GB$1,499Fits17GB headroomest. 48.3 tok/s
- Mac mini, M5 Pro, 24GB$1,699Fits9GB headroomest. 87.2 tok/s
- Mac mini, M5 Pro, 48GB$2,499Fits32GB headroomest. 87.2 tok/s
- Mac Studio, M5 Max, 36GB$2,499Fits21GB headroomest. 174 tok/s
- Mac mini, M5 Pro, 64GB$2,899Fits47GB headroomest. 87.2 tok/s
- GeForce RTX 5090, 32GB$2,999.99+hostFits19GB headroomest. 509 tok/s
- Mac Studio, M5 Max, 48GB$3,099Fits32GB headroomest. 174 tok/s
- Mac Studio, M5 Max, 64GB$3,499Fits47GB headroomest. 174 tok/s
- Fits108GB headroomest. 77.6 tok/s
- Fits108GB headroomest. 77.6 tok/s
- Fits108GB headroomest. 174 tok/s
- Fits76GB headroomest. 341 tok/s
- Fits236GB headroomest. 341 tok/s
- Fits492GB headroomest. 341 tok/s
| Config | Memory | Price | $/GB | Fit | Headroom | Est. speed |
|---|---|---|---|---|---|---|
| Mac mini, M6 | 16GB | $1,099 | $69 | Tight | 1GB | 48.3 tok/s |
| Mac mini, M6 | 24GB | $1,299 | $54 | Fits | 9GB | 48.3 tok/s |
| Mac mini, M6 | 32GB | $1,499 | $47 | Fits | 17GB | 48.3 tok/s |
| Mac mini, M5 Pro | 24GB | $1,699 | $71 | Fits | 9GB | 87.2 tok/s |
| Mac mini, M5 Pro | 48GB | $2,499 | $52 | Fits | 32GB | 87.2 tok/s |
| Mac Studio, M5 Max | 36GB | $2,499 | $69 | Fits | 21GB | 174 tok/s |
| Mac mini, M5 Pro | 64GB | $2,899 | $45 | Fits | 47GB | 87.2 tok/s |
| GeForce RTX 5090 | 32GB | $2,999.99+host | $94 | Fits | 19GB | 509 tok/s |
| Mac Studio, M5 Max | 48GB | $3,099 | $65 | Fits | 32GB | 174 tok/s |
| Mac Studio, M5 Max | 64GB | $3,499 | $55 | Fits | 47GB | 174 tok/s |
| Ascent GX10, GB10 Grace Blackwell | 128GB | $3,999.99 | $31 | Fits | 108GB | 77.6 tok/s |
| DGX Spark, GB10 Grace Blackwell | 128GB | $4,699 | $37 | Fits | 108GB | 77.6 tok/s |
| Mac Studio, M5 Max | 128GB | $5,099 | $40 | Fits | 108GB | 174 tok/s |
| Mac Studio, M5 Ultra | 96GB | $5,499 | $57 | Fits | 76GB | 341 tok/s |
| Mac Studio, M5 Ultra | 256GB | $9,499 | $37 | Fits | 236GB | 341 tok/s |
| Mac Studio, M5 Ultra | 512GB | — | — | Fits | 492GB | 341 tok/s |
GeForce RTX 5090 is a card, not a computer: every price shown for it is the cheapest listing we found plus $1,000 for the minimum host machine.
memory = 21B × 0.489 B/param (3.91 bits) + 0.049 GB/1K × 8K K/V + 1.2 GB runtime
speed = bandwidth × 0.5 ÷ (3.6B active × 0.489 B/param) = 1.8 GB read per token
usable = memory − operating system reserve (8% on a Mac, 6% on a DGX Spark, 4% on a card)
Bandwidth is the vendor’s published figure. Every speed is an estimate and says so. Methodology.
What else Mac mini M6 24GB runs
- Qwen3-32BQ3_K_M, est. 7.1 tok/s
- Gemma-3-27BQ4_K_M, est. 6.8 tok/s
- gpt-oss-20bQ6_K, est. 28.8 tok/s
Best quant that still fits at 8K context, largest model first. Full list on the Config page.
Bandwidth 170GB/s, usable memory 21GB of 24GB.