How much VRAM does gpt-oss-120b need?

At Q4_K_M with 8K context: 71.5GB weights + 0.6GB KV cache + 1.5GB runtime overhead = 73.6GB total. Mac Studio M5 Ultra (512GB) offers 384.0GB usable VRAM, so the verdict is: Comfortable.

What is the best quantization for gpt-oss-120b on Mac Studio M5 Ultra (512GB)?

FP16 — the highest tier that still fits within 384.0GB usable VRAM at 8K context. Lower tiers (Q3/Q2) fit too but cost noticeable quality.

How fast does gpt-oss-120b run on Mac Studio M5 Ultra (512GB)?

About 88 tokens/s at FP16 (theoretical estimate, ±30% in real-world use).

Can Mac Studio M5 Ultra (512GB) run gpt-oss-120b?

✅ Comfortable

gpt-oss-120b @ Q4_K_M · 8K context

Needed

73.6 GB

Usable

384.0 GB

VRAM breakdown

Weights (Q4_K_M)	71.5 GB
KV cache (8K context)	0.6 GB
Runtime overhead	1.5 GB
Total	73.6 GB

Computed at 8K context with fp16 KV cache. Longer contexts need more VRAM — use the VRAM calculator for other settings.

Recommended quantization

FP16

Estimated generation speed

~88 tok/s @ FP16

Theoretical estimates based on published architecture data and measured GGUF sizes; real-world speed varies ±30%.

Smaller models this GPU runs well

Other GPUs that run gpt-oss-120b

FAQ

How much VRAM does gpt-oss-120b need?: At Q4_K_M with 8K context: 71.5GB weights + 0.6GB KV cache + 1.5GB runtime overhead = 73.6GB total. Mac Studio M5 Ultra (512GB) offers 384.0GB usable VRAM, so the verdict is: Comfortable.
What is the best quantization for gpt-oss-120b on Mac Studio M5 Ultra (512GB)?: FP16 — the highest tier that still fits within 384.0GB usable VRAM at 8K context. Lower tiers (Q3/Q2) fit too but cost noticeable quality.
How fast does gpt-oss-120b run on Mac Studio M5 Ultra (512GB)?: About 88 tokens/s at FP16 (theoretical estimate, ±30% in real-world use).

Check another combination in the GPU Checker →

Data verified 2026-11-01