How much memory does gpt-oss-20b at BF16 need?

44GB. That is 42GB of weights at 16 bits per weight, 0.4GB of K/V cache at 8K context, and 1.2GB of runtime overhead.

Which machine is cheapest for gpt-oss-20b at BF16?

Mac mini, M5 Pro, 64GB, $2,899 at Apple. It fits with 15GB of headroom and decodes at an est. 21.3 tok/s.

Cheapest box to run gpt-oss-20b at BF16

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.

Fit finderprices checked 19 minutes ago

Quant

Context

gpt-oss-20b at BF16 needs 44GB.

Cheapest Config that fits: Mac mini, M5 Pro, 64GB. Fits, 15GB headroom. Est. 21.3 tok/s.

Best offer today: $2,899 at Apple — in stock, checked 19 minutes ago.

Price history: 47 days recorded.

Offers — Mac mini, M5 Pro, 64GB, 512GB

B&H PhotoPreorder, ships about 22 Sep. Below Apple's own price for the same build, which is unusual enough that we are re-checking both numbers.out of stockchecked 46 days ago$2,699
Applelist pricein stockchecked 19 minutes ago$2,899

Mac mini, M6, 16GB$1,099
Won't fit−31GB headroom
Mac mini, M6, 24GB$1,299
Won't fit−23GB headroom
Mac mini, M6, 32GB$1,499
Won't fit−15GB headroom
Mac mini, M5 Pro, 24GB$1,699
Won't fit−23GB headroom
Mac mini, M5 Pro, 48GB$2,499
Tight1GB headroomest. 21.3 tok/s
Mac Studio, M5 Max, 36GB$2,499
Won't fit−11GB headroom
Mac mini, M5 Pro, 64GB$2,899
Fits15GB headroomest. 21.3 tok/s
GeForce RTX 5090, 32GB$2,999.99+host
Won't fit−13GB headroom
Mac Studio, M4 Max, 64GB, refurbished$3,049
Fits15GB headroomest. 37.9 tok/s
Mac Studio, M5 Max, 48GB$3,099
Tight1GB headroomest. 42.6 tok/s
Mac Studio, M5 Max, 64GB$3,499
Fits15GB headroomest. 42.6 tok/s
Ascent GX10, GB10 Grace Blackwell, 128GB$3,999.99
Fits77GB headroomest. 19.0 tok/s
Mac Studio, M3 Ultra, 96GB, refurbished$4,329
Fits45GB headroomest. 56.9 tok/s
Mac Studio, M4 Max, 128GB, refurbished$4,409
Fits76GB headroomest. 37.9 tok/s
DGX Spark, GB10 Grace Blackwell, 128GB$4,699
Fits77GB headroomest. 19.0 tok/s
Mac Studio, M5 Max, 128GB$5,099
Fits76GB headroomest. 42.6 tok/s
Mac Studio, M5 Ultra, 96GB$5,499
Fits45GB headroomest. 83.3 tok/s
Mac Studio, M3 Ultra, 256GB, refurbished$7,729
Fits204GB headroomest. 56.9 tok/s
Mac Studio, M5 Ultra, 256GB$9,499
Fits204GB headroomest. 83.3 tok/s
Mac Studio, M5 Ultra, 512GB—
Fits460GB headroomest. 83.3 tok/s

Config	Memory	Price	Fit	Headroom	Est. speed
Mac mini, M6	16GB	$1,099	Won't fit	−31GB	—
Mac mini, M6	24GB	$1,299	Won't fit	−23GB	—
Mac mini, M6	32GB	$1,499	Won't fit	−15GB	—
Mac mini, M5 Pro	24GB	$1,699	Won't fit	−23GB	—
Mac mini, M5 Pro	48GB	$2,499	Tight	1GB	21.3 tok/s
Mac Studio, M5 Max	36GB	$2,499	Won't fit	−11GB	—
Mac mini, M5 Pro	64GB	$2,899	Fits	15GB	21.3 tok/s
GeForce RTX 5090	32GB	$2,999.99+host	Won't fit	−13GB	—
Mac Studio, M4 Max, refurb	64GB	$3,049	Fits	15GB	37.9 tok/s
Mac Studio, M5 Max	48GB	$3,099	Tight	1GB	42.6 tok/s
Mac Studio, M5 Max	64GB	$3,499	Fits	15GB	42.6 tok/s
Ascent GX10, GB10 Grace Blackwell	128GB	$3,999.99	Fits	77GB	19.0 tok/s
Mac Studio, M3 Ultra, refurb	96GB	$4,329	Fits	45GB	56.9 tok/s
Mac Studio, M4 Max, refurb	128GB	$4,409	Fits	76GB	37.9 tok/s
DGX Spark, GB10 Grace Blackwell	128GB	$4,699	Fits	77GB	19.0 tok/s
Mac Studio, M5 Max	128GB	$5,099	Fits	76GB	42.6 tok/s
Mac Studio, M5 Ultra	96GB	$5,499	Fits	45GB	83.3 tok/s
Mac Studio, M3 Ultra, refurb	256GB	$7,729	Fits	204GB	56.9 tok/s
Mac Studio, M5 Ultra	256GB	$9,499	Fits	204GB	83.3 tok/s
Mac Studio, M5 Ultra	512GB	—	Fits	460GB	83.3 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 × 2.000 B/param (16 bits) + 0.049 GB/1K × 8K K/V + 1.2 GB runtime
speed = bandwidth × 0.5 ÷ (3.6B active × 2.000 B/param) = 7.2 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 M5 Pro 64GB runs

Llama-4-Scout-17B-16EQ3_K_M, est. 18.5 tok/s
Qwen3-Coder-NextQ5_K_M, est. 72.2 tok/s
Llama-3.3-70BQ5_K_M, est. 4.0 tok/s
Qwen3.5-35B-A3BQ8_0, est. 48.2 tok/s
Qwen3-32BQ8_0, est. 5.9 tok/s
Gemma-4-31BQ8_0, est. 6.1 tok/s

Best quant that still fits at 8K context, largest model first. Full list on the Config page.

Bandwidth 307GB/s, usable memory 59GB of 64GB.