buyaigear
Prices checked 10 min ago

Apple · M3 Ultra · 512 GB · Qwen3.8 27B · Q4_K_M: ~30 tok/s

Apple Mac Studio M3 Ultra, 512 GB

The only desktop that holds a 400B-class model: 512 GB of unified memory at 819 GB/s. Decode is quick for the size; prefill is the weak point, since there are no tensor cores and the Neural Engine sits unused by llama.cpp.

Lowest price right now

$19,300

eBay · free shipping · $37 per GB · observed 2026-10-14 22:19 UTC

See it on eBay

Where to buy

VendorPriceConditionObserved
eBay
cheapest active listing		 $19,300
free shipping		 new and used 2026-10-14 22:19 UTC Go to eBay

On eBay now

10 listings passed our filters at the last check. Cheapest delivered first. Seller reputation is feedback count and positive rate.

All 10 listings · Open the search on eBay

What it runs

For each model, grouped by vendor with the newest first: the largest quant that fits with 8K of context and the estimated speed for one user. The model you picked on the ladder is highlighted. Estimates from the spec sheet, not measurements: see how we estimate.

ModelQuantNeeds at 8KFitMax contextDecode tok/sPrefill tok/s
DeepSeek
DeepSeek V4.1 Flash 552B-A16B Sep 2026Q6_K454 GBfits128K+~37~220
DeepSeek V4 Flash 284B-A13B Jul 2026Q8_0303 GBfits128K+~36~270
DeepSeek V3.1 671B-A37B Aug 2025Q5_K_M478 GBfits128K+~19~96
Gemma
Gemma 4 12B May 2026Q8_014 GBfits128K+~39~300
Gemma 4 31B Mar 2026Q8_035 GBfits128K+~15~120
Gemma 4 26B-A4B Mar 2026Q8_028 GBfits128K+~120~930
Gemma 3 27B Mar 2025Q8_031 GBfits128K+~17~130
GLM
GLM-5.3 753B-A40B Aug 2026Q4_K_M462 GBfits128K+~20~89
GLM-5.3 Flash 320B-A18B Aug 2026Q8_0341 GBfits128K+~26~200
GLM-4.6 355B-A32B Sep 2025Q8_0381 GBfits128K+~14~110
GLM-4.5 Air 106B-A12B Jul 2025Q8_0115 GBfits128K+~39~300
gpt-oss
gpt-oss-120b Aug 2025Q8_0126 GBfits128K+~91~700
gpt-oss-20b Aug 2025Q8_024 GBfits128K+~130~990
Llama
Llama 4 Scout 109B-A17B Apr 2025Q8_0119 GBfits128K+~27~210
Llama 3.3 70B Dec 2024Q8_079 GBfits128K+~6.6~50
Llama 3.1 8B Jul 2024Q8_011 GBfits128K+~58~440
MiMo
MiMo V2.6 Pro 1.02T-A42B Sep 2026Q3_K_M502 GBfits128K+~24~85
MiMo V2.6 Flash 309B-A15B Sep 2026Q8_0329 GBfits128K+~31~240
Mistral
Mistral Medium 3.5 128B Mar 2026Q8_0140 GBfits128K+~3.6~28
Mistral Small 4 119B-A6.5B Mar 2026Q8_0132 GBfits128K+~71~550
Mistral Small 3.2 24B Jun 2025Q8_028 GBfits128K+~19~150
Qwen
Qwen3.8 Flash-Next 125B-A6B Aug 2026Q8_0134 GBfits128K+~77~590
Qwen3.8 27B Aug 2026 (your pick)Q8_031 GBfits128K+~17~130
Qwen3.6 27B Apr 2026Q8_031 GBfits128K+~17~130
Qwen3.6 35B-A3B Apr 2026Q8_039 GBfits128K+~150~1,200
Qwen3 Coder 480B-A35B Jul 2025Q6_K397 GBfits128K+~17~100
Qwen3 235B-A22B Jul 2025Q8_0252 GBfits128K+~21~160
Qwen3 32B Apr 2025Q8_038 GBfits128K+~14~110

Specifications

Memory
512 GB LPDDR5 unified
Memory bandwidth
819 GB/s
Bus width
1024-bit
FP32
28.4 TFLOPS
FP16 (dense)
28 TFLOPS
Form factor
Desktop box, 197 x 197 x 95 mm
CPU
32-core CPU, 80-core GPU, 32-core Neural Engine
Architecture
M3 Ultra
Launched
2025
MSRP
$9,499

Source: manufacturer specifications.