Cheapest box to run Qwen3-235B-A22B at BF16
Alibaba’s Qwen3-235B-A22B is 235B parameters, 22B 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.
Qwen3-235B-A22B at BF16 needs 473GB.
Cheapest Config that fits: Mac Studio, M5 Ultra, 512GB. Tight, only 31GB spare. Est. 13.6 tok/s.
No price yet: Ships late October 2026. Apple has not published a price.
Collecting price history — 1 of 90 days. No timing verdict yet.
Offers — Mac Studio, M5 Ultra, 512GB, 1TB
Ships late October 2026. Apple has not published a price.
- Mac mini, M6, 16GB$1,099Won't fit−460GB headroom
- Mac mini, M6, 24GB$1,299Won't fit−452GB headroom
- Mac mini, M6, 32GB$1,499Won't fit−444GB headroom
- Mac mini, M5 Pro, 24GB$1,699Won't fit−452GB headroom
- Mac mini, M5 Pro, 48GB$2,499Won't fit−429GB headroom
- Mac Studio, M5 Max, 36GB$2,499Won't fit−440GB headroom
- Mac mini, M5 Pro, 64GB$2,899Won't fit−414GB headroom
- GeForce RTX 5090, 32GB$2,999.99+hostWon't fit−442GB headroom
- Mac Studio, M5 Max, 48GB$3,099Won't fit−429GB headroom
- Mac Studio, M5 Max, 64GB$3,499Won't fit−414GB headroom
- Won't fit−352GB headroom
- Won't fit−352GB headroom
- Won't fit−353GB headroom
- Won't fit−384GB headroom
- Won't fit−225GB headroom
- Tight31GB headroomest. 13.6 tok/s
| Config | Memory | Price | $/GB | Fit | Headroom | Est. speed |
|---|---|---|---|---|---|---|
| Mac mini, M6 | 16GB | $1,099 | $69 | Won't fit | −460GB | — |
| Mac mini, M6 | 24GB | $1,299 | $54 | Won't fit | −452GB | — |
| Mac mini, M6 | 32GB | $1,499 | $47 | Won't fit | −444GB | — |
| Mac mini, M5 Pro | 24GB | $1,699 | $71 | Won't fit | −452GB | — |
| Mac mini, M5 Pro | 48GB | $2,499 | $52 | Won't fit | −429GB | — |
| Mac Studio, M5 Max | 36GB | $2,499 | $69 | Won't fit | −440GB | — |
| Mac mini, M5 Pro | 64GB | $2,899 | $45 | Won't fit | −414GB | — |
| GeForce RTX 5090 | 32GB | $2,999.99+host | $94 | Won't fit | −442GB | — |
| Mac Studio, M5 Max | 48GB | $3,099 | $65 | Won't fit | −429GB | — |
| Mac Studio, M5 Max | 64GB | $3,499 | $55 | Won't fit | −414GB | — |
| Ascent GX10, GB10 Grace Blackwell | 128GB | $3,999.99 | $31 | Won't fit | −352GB | — |
| DGX Spark, GB10 Grace Blackwell | 128GB | $4,699 | $37 | Won't fit | −352GB | — |
| Mac Studio, M5 Max | 128GB | $5,099 | $40 | Won't fit | −353GB | — |
| Mac Studio, M5 Ultra | 96GB | $5,499 | $57 | Won't fit | −384GB | — |
| Mac Studio, M5 Ultra | 256GB | $9,499 | $37 | Won't fit | −225GB | — |
| Mac Studio, M5 Ultra | 512GB | — | — | Tight | 31GB | 13.6 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 = 235B × 2.000 B/param (16 bits) + 0.193 GB/1K × 8K K/V + 1.2 GB runtime
speed = bandwidth × 0.5 ÷ (22B active × 2.000 B/param) = 44.0 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 Studio M5 Ultra 512GB runs
- DeepSeek-V3Q5_K_M, est. 22.9 tok/s
- Llama-4-Maverick-17B-128EQ8_0, est. 33.2 tok/s
- Qwen3-235B-A22BBF16, est. 13.6 tok/s
- Mistral-Large-2411BF16, est. 3.2 tok/s
- gpt-oss-120bBF16, est. 58.8 tok/s
- Llama-4-Scout-17B-16EBF16, est. 17.6 tok/s
Best quant that still fits at 8K context, largest model first. Full list on the Config page.
Bandwidth 1.2TB/s, usable memory 504GB of 512GB.