RunsOnMy ModelsGPUs & MacsBy VRAMAPI pricesVRAM calculatorNew

Can I run gpt oss 20b locally?

gpt oss 20b by openai has 21B parameters (4.2B active per token, 32 experts, 4 used per token). At 4-bit with an 8K context it needs about 11.8 GB of GPU memory. It fits on 48 of 65 GPUs and Macs we track. See all models that fit in 16 GB of VRAM.

11.8 GBQ4_K_M · 8K context
RTX 4060 Ti 16GBsmallest consumer GPU that fits
M2 24GBsmallest Mac that fits
128Kmax context
GQA + sliding windowattention type

Memory needed by quant and context length

GB of GPU memory = weights + KV cache (FP16) + runtime overhead. Weights use real file sizes from unsloth/gpt-oss-20b-GGUF where available (marked •), otherwise parameter count × bits per weight. How we calculate.

QuantQuality4K ctx8K ctx32K ctx128K ctx
2-bit (Q2_K) •rough11.6 GB11.7 GB12.2 GB14.5 GB
3-bit (Q3_K_M) •usable11.6 GB11.7 GB12.3 GB14.5 GB
4-bit (Q4_K_M) •good, most popular11.7 GB11.8 GB12.4 GB14.6 GB
5-bit (Q5_K_M) •very good11.8 GB11.9 GB12.4 GB14.7 GB
6-bit (Q6_K) •near lossless12.1 GB12.2 GB12.8 GB15.0 GB
8-bit (Q8_0) •effectively lossless12.2 GB12.3 GB12.8 GB15.1 GB
16-bit (F16/BF16) •original weights13.7 GB13.8 GB14.4 GB16.6 GB

Which GPUs and Macs run gpt oss 20b?

Best quality quant that fits at 8K context. "Tight" = within 10% of usable memory. Speed is a ceiling from memory bandwidth (real speed is typically 50–80% of it).

HardwareMemoryFits?Best quantNeedsMax tok/s (est.)
RTX 3060 12GB12 GBNo———details
RTX 4060 Ti 16GB16 GBFitsQ8_012.3 GB119details
RTX 309024 GBFitsF1613.8 GB338details
RTX 409024 GBFitsF1613.8 GB365details
RTX 509032 GBFitsF1613.8 GB648details
RTX 5070 Ti16 GBFitsQ8_012.3 GB369details
RX 7900 XTX24 GBFitsF1613.8 GB347details
Mac M4 Pro 48GB48 GBFitsF1613.8 GB99details
Mac M4 Max 128GB128 GBFitsF1613.8 GB197details
Mac M3 Ultra 512GB512 GBFitsF1613.8 GB296details
DGX Spark128 GBFitsF1613.8 GB99details
RTX 407012 GBNo———details
RTX 5060 Ti 16GB16 GBFitsQ8_012.3 GB185details
RTX 40608 GBNo———details
Arc B58012 GBNo———details
RTX 3060 Ti8 GBNo———details
RTX 30708 GBNo———details
RTX 3080 10GB10 GBNo———details
RTX 3080 12GB12 GBNo———details
RTX 3090 Ti24 GBFitsF1613.8 GB365details
RTX 4060 Ti 8GB8 GBNo———details
RTX 4070 Super12 GBNo———details
RTX 4070 Ti12 GBNo———details
RTX 4070 Ti Super16 GBFitsQ8_012.3 GB277details
RTX 408016 GBFitsQ8_012.3 GB295details
RTX 4080 Super16 GBFitsQ8_012.3 GB303details
RTX 50608 GBNo———details
RTX 507012 GBNo———details
RTX 508016 GBFitsQ8_012.3 GB395details
2x RTX 30902×24 GBFitsF1613.8 GB338details
2x RTX 40902×24 GBFitsF1613.8 GB365details
2x RTX 50902×32 GBFitsF1613.8 GB648details
RTX A600048 GBFitsF1613.8 GB278details
RTX 6000 Ada48 GBFitsF1613.8 GB347details
RTX PRO 6000 Blackwell96 GBFitsF1613.8 GB648details
L40S48 GBFitsF1613.8 GB312details
A100 40GB40 GBFitsF1613.8 GB562details
A100 80GB80 GBFitsF1613.8 GB737details
H100 PCIe80 GBFitsF1613.8 GB723details
H100 SXM80 GBFitsF1613.8 GB1211details
H200141 GBFitsF1613.8 GB1736details
RX 76008 GBNo———details
RX 7700 XT12 GBNo———details
RX 7800 XT16 GBFitsQ8_012.3 GB257details
RX 7900 GRE16 GBFitsQ8_012.3 GB237details
RX 7900 XT20 GBFitsF1613.8 GB289details
RX 907016 GBFitsQ8_012.3 GB264details
RX 9070 XT16 GBFitsQ8_012.3 GB264details
Ryzen AI Max+ 395 128GB128 GBFitsF1613.8 GB93details
Arc A770 16GB16 GBFitsQ8_012.3 GB231details
Mac M1 16GB16 GBNo———details
Mac M2 24GB24 GBFitsF1613.8 GB36details
Mac M3 24GB24 GBFitsF1613.8 GB36details
Mac M4 16GB16 GBNo———details
Mac M4 32GB32 GBFitsF1613.8 GB43details
Mac M1 Pro 32GB32 GBFitsF1613.8 GB72details
Mac M3 Pro 36GB36 GBFitsF1613.8 GB54details
Mac M4 Pro 24GB24 GBFitsF1613.8 GB99details
Mac M4 Pro 64GB64 GBFitsF1613.8 GB99details
Mac M1 Max 64GB64 GBFitsF1613.8 GB145details
Mac M2 Max 96GB96 GBFitsF1613.8 GB145details
Mac M3 Max 128GB128 GBFitsF1613.8 GB145details
Mac M4 Max 64GB64 GBFitsF1613.8 GB197details
Mac M2 Ultra 192GB192 GBFitsF1613.8 GB289details
Mac M3 Ultra 256GB256 GBFitsF1613.8 GB296details

Prefer an API?

OpenAI: gpt-oss-20b is on OpenRouter at $0.03 per 1M input tokens and $0.13 per 1M output tokens. Price details and history →

Questions

How much VRAM does gpt oss 20b need?

About 11.8 GB at 4-bit (Q4_K_M) with an 8K context: 10.8 GB of weights (real GGUF file size), 0.2 GB of KV cache and 0.8 GB of runtime overhead. At 8-bit it needs about 12.3 GB.

What is the smallest GPU that runs gpt oss 20b?

At Q4_K_M and 8K context, the smallest single consumer GPU in our table that fits it is the RTX 4060 Ti 16GB (16 GB).

Can I run gpt oss 20b on a Mac?

Yes. At Q4_K_M with 8K context it fits on a Mac M2 24GB or larger, using the default macOS GPU memory limit.

Similar models

Same family or a similar parameter count, with the VRAM each needs at Q4_K_M and 8K context.

gpt oss safeguard 20b

22B · 11.8 GB at 4-bit

gpt oss 120b

117B · 59.5 GB at 4-bit

gpt oss safeguard 120b

117B · 59.5 GB at 4-bit

ERNIE 4.5 21B A3B Thinking

22B · 13.5 GB at 4-bit

ERNIE 4.5 21B A3B PT

22B · 13.5 GB at 4-bit

Groot 19B A3B 261023

19B · 12.7 GB at 4-bit

Model facts

Hugging Face repoopenai/gpt-oss-20b
Parameters20,914,757,184 (21B) · ~4.2B active (computed from config)
Layers / hidden size24 / 2880
AttentionGQA + sliding window; 64 heads, 8 KV heads, head dim 64; sliding window 128
KV cache per token (FP16)48.0 KiB (sliding layers capped at the window)
Max context131,072 tokens
Licenseapache-2.0
Released on HFAug 4, 2025
Downloads (30d) / likes5,575,731 / 5,320

Sources: Hugging Face model API · config.json (openai/gpt-oss-20b) · GGUF files · GPU specs from vendor pages (list). Data fetched Nov 17, 2026. Spot an error? Report it.