Cheapest box to run Qwen3.5-35B-A3B at BF16
Alibaba’s Qwen3.5-35B-A3B is 35B parameters, 3B 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.
Qwen3.5-35B-A3B at BF16 needs 71GB.
Cheapest Config that fits: Ascent GX10, GB10 Grace Blackwell, 128GB. Fits, 49GB headroom. Est. 22.8 tok/s.
Best offer today: $3,999.99 at Best Buy — in stock, checked 33 days ago.
Price history: 34 days recorded.
A hybrid: only 10 of 40 layers are full attention. The Gated DeltaNet layers keep a constant-size state and cache nothing per token, so the K/V figure counts the full-attention layers alone.
Offers — Ascent GX10, GB10 Grace Blackwell, 128GB, 1TB
- Best BuyRead on 28 Aug 2026 and not re-read since: Best Buy blocks our automated checks, so this number is old and we say so rather than imply it is current. It was $1,999 under ASUS's own store once ASUS raised its price in September.in stockchecked 33 days ago$3,999.99
- ASUSlist pricein stockchecked 2 hours ago$5,999
- Mac mini, M6, 16GB$1,099Won't fit−58GB headroom
- Mac mini, M6, 24GB$1,299Won't fit−50GB headroom
- Mac mini, M6, 32GB$1,499Won't fit−42GB headroom
- Mac mini, M5 Pro, 24GB$1,699Won't fit−50GB headroom
- Mac mini, M5 Pro, 48GB$2,499Won't fit−27GB headroom
- Mac Studio, M5 Max, 36GB$2,499Won't fit−38GB headroom
- Mac mini, M5 Pro, 64GB$2,899Won't fit−12GB headroom
- GeForce RTX 5090, 32GB$2,999.99+hostWon't fit−41GB headroom
- Won't fit−12GB headroom
- Mac Studio, M5 Max, 48GB$3,099Won't fit−27GB headroom
- Mac Studio, M5 Max, 64GB$3,499Won't fit−12GB headroom
- Fits49GB headroomest. 22.8 tok/s
- Fits17GB headroomest. 68.3 tok/s
- Fits49GB headroomest. 45.5 tok/s
- Fits49GB headroomest. 22.8 tok/s
- Fits49GB headroomest. 51.2 tok/s
- Fits17GB headroomest. 100 tok/s
- Fits177GB headroomest. 68.3 tok/s
- Fits177GB headroomest. 100 tok/s
- Fits433GB headroomest. 100 tok/s
| Config | Memory | Price | $/GB | Fit | Headroom | Est. speed |
|---|---|---|---|---|---|---|
| Mac mini, M6 | 16GB | $1,099 | $69 | Won't fit | −58GB | — |
| Mac mini, M6 | 24GB | $1,299 | $54 | Won't fit | −50GB | — |
| Mac mini, M6 | 32GB | $1,499 | $47 | Won't fit | −42GB | — |
| Mac mini, M5 Pro | 24GB | $1,699 | $71 | Won't fit | −50GB | — |
| Mac mini, M5 Pro | 48GB | $2,499 | $52 | Won't fit | −27GB | — |
| Mac Studio, M5 Max | 36GB | $2,499 | $69 | Won't fit | −38GB | — |
| Mac mini, M5 Pro | 64GB | $2,899 | $45 | Won't fit | −12GB | — |
| GeForce RTX 5090 | 32GB | $2,999.99+host | $94 | Won't fit | −41GB | — |
| Mac Studio, M4 Max, refurb | 64GB | $3,049 | $48 | Won't fit | −12GB | — |
| Mac Studio, M5 Max | 48GB | $3,099 | $65 | Won't fit | −27GB | — |
| Mac Studio, M5 Max | 64GB | $3,499 | $55 | Won't fit | −12GB | — |
| Ascent GX10, GB10 Grace Blackwell | 128GB | $3,999.99 | $31 | Fits | 49GB | 22.8 tok/s |
| Mac Studio, M3 Ultra, refurb | 96GB | $4,329 | $45 | Fits | 17GB | 68.3 tok/s |
| Mac Studio, M4 Max, refurb | 128GB | $4,409 | $34 | Fits | 49GB | 45.5 tok/s |
| DGX Spark, GB10 Grace Blackwell | 128GB | $4,699 | $37 | Fits | 49GB | 22.8 tok/s |
| Mac Studio, M5 Max | 128GB | $5,099 | $40 | Fits | 49GB | 51.2 tok/s |
| Mac Studio, M5 Ultra | 96GB | $5,499 | $57 | Fits | 17GB | 100 tok/s |
| Mac Studio, M3 Ultra, refurb | 256GB | $7,729 | $30 | Fits | 177GB | 68.3 tok/s |
| Mac Studio, M5 Ultra | 256GB | $9,499 | $37 | Fits | 177GB | 100 tok/s |
| Mac Studio, M5 Ultra | 512GB | — | — | Fits | 433GB | 100 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 = 35B × 2.000 B/param (16 bits) + 0.021 GB/1K × 8K K/V + 1.2 GB runtime
speed = bandwidth × 0.5 ÷ (3B active × 2.000 B/param) = 6.0 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 Ascent GX10 GB10 Grace Blackwell 128GB runs
- Qwen3-235B-A22BQ3_K_M, est. 12.7 tok/s
- MiniMax-M2Q3_K_M, est. 27.9 tok/s
- Mistral-Large-2411Q6_K, est. 1.8 tok/s
- gpt-oss-120bQ6_K, est. 32.6 tok/s
- Llama-4-Scout-17B-16EQ8_0, est. 7.6 tok/s
- Qwen3-Coder-NextQ8_0, est. 42.8 tok/s
Best quant that still fits at 8K context, largest model first. Full list on the Config page.
Bandwidth 273GB/s, usable memory 120GB of 128GB.