LLMBOTTLENECK.COM
OpenAI / gpt_oss

gpt-oss-safeguard-20b

21B total · 3.6B active parameters, routing 4 of 32 experts per token.

OpenAI's official card declares 21B total parameters and 3.6B active parameters. The U8 safetensors total is a packed checkpoint representation and is not shown as an unqualified logical count. Open the official parameter note ↗

Open in the calculator →

Architecturepublished data
Quick answer
gpt-oss-safeguard-20b needs about 13.8 GB at Q4_K_M — the format most people download — with 8,192 tokens of context. It fits entirely on 17 of 18 common devices, starting with the AMD Radeon™ RX 9070 XT (16 GB). What else fits in 16 GB. No hardware for it? Rent a GPU that runs it, priced live.

Sized at 8,192 tokens of context with the whole model in device memory, across 18 common devices. Speeds are estimates, not benchmarks.

Find the AMD Radeon™ RX 9070 XT, the smallest card listed here that holds it: Amazon ↗ · eBay (new and used) ↗Store links may pay us a commission. They never decide which card is suggested — the memory arithmetic does.

Weight-file size by format

FormatSizeRangeBasis
FP1641.8 GB40.7 GB – 42.2 GBreconstructed size
Q8_022.2 GB21.6 GB – 22.4 GBreconstructed size
Q6_K17.2 GB16.7 GB – 17.3 GBreconstructed size
Q5_K_M14.9 GB14.4 GB – 15.0 GBreconstructed size
Q5_014.5 GB14.0 GB – 14.6 GBreconstructed size
Q4_K_M12.8 GB12.3 GB – 12.9 GBreconstructed size
Q4_011.9 GB11.4 GB – 12.1 GBreconstructed size
Q3_K_M10.5 GB9.70 GB – 11.3 GBreconstructed size
Q2_K7.81 GB7.33 GB – 8.16 GBreconstructed size

A file size is not the memory a run needs: the KV cache and the runtime reserve come on top, and the calculator adds both.

What the labels mean
published data
Read from a published source — a file's byte count, a model's configuration or a manufacturer's specification — or exact arithmetic on such values. Not a measurement on a machine.
reconstructed size
Weight size reconstructed from the pinned architecture, because no published file exists.
size range
Only a lower and an upper bound are claimed for this weight size.

Hardware ladder

Q4_K_M at 8,192 tokens · 17 of 18 devices hold it entirely
DeviceMemoryVerdictDecodeCalculator
NVIDIA GeForce RTX 507013.8 GB of 12 GB4 layers on system RAM~91 tok/s with offloadOpen →
AMD Radeon™ RX 9070 XT13.8 GB of 16 GBfits~199 tok/sOpen →
NVIDIA GeForce RTX 408013.8 GB of 16 GBfits~195 tok/sOpen →
NVIDIA GeForce RTX 508013.8 GB of 16 GBfits~261 tok/sOpen →
NVIDIA GeForce RTX 5070 Ti13.8 GB of 16 GBfits~244 tok/sOpen →
NVIDIA GeForce RTX 5060 Ti 16GB13.8 GB of 16 GBfits~122 tok/sOpen →
NVIDIA GeForce RTX 409013.8 GB of 24 GBfits~274 tok/sOpen →
AMD Radeon™ RX 7900 XTX13.8 GB of 24 GBfits~299 tok/sOpen →
NVIDIA GeForce RTX 309013.8 GB of 24 GBfits~254 tok/sOpen →
NVIDIA GeForce RTX 509013.8 GB of 32 GBfits~487 tok/sOpen →
NVIDIA RTX 6000 Ada Generation13.8 GB of 48 GBfits~261 tok/sOpen →
Apple M4 Pro13.8 GB of 64 GBfits~65 tok/sOpen →
Apple M5 Pro13.8 GB of 64 GBfits~70 tok/sOpen →
AMD Ryzen AI Max+ 395 with Radeon 8060S13.8 GB of 128 GBfits~80 tok/sOpen →
Apple M3 Max13.8 GB of 128 GBfits~81 tok/sOpen →
Apple M4 Max13.8 GB of 128 GBfits~95 tok/sOpen →
NVIDIA DGX Spark13.8 GB of 128 GBfits~74 tok/sOpen →
Apple M2 Ultra13.8 GB of 192 GBfits~111 tok/sOpen →

Decode is a calibrated estimate from the published calibration; capacity is the manufacturer's published ceiling, not guaranteed free memory. Fitting in memory is not the same as loading: whether the runtime and version you have supports this architecture and format on that machine has not been tested here. Offloaded rows assume there is enough system RAM for the overflow — the calculator checks that against the RAM you declare. Each “Open” link keeps this model, format and context.

Run it in the cloud

Rent a GPU that runs gpt-oss-safeguard-20b

Every rentable machine that holds the whole model, from Vast.ai and RunPod, with this site's speed estimate for each and the provider's own price, read live. Cheapest first; sort by cost per token or speed instead, or switch the format.

Prices · 18:30 UTC, 22 Sept
27 configurations in stock · 6 out of stock
MachineSpeedPer hourPer M tokensWhereRent
1× RTX 4060 Ti 16GB16 GBcheapest~78 tok/sFaster than you read$0.090$0.32Vast.aiMarketplace · 99.4% reliableRent on Vast.ai
2× RTX 306024 GB~195 tok/sFaster than you read$0.11$0.15Vast.aiMarketplace · 98.0% reliableRent on Vast.ai
1× RTX 309024 GBrecommendedbest value~254 tok/sFaster than you read$0.12$0.13Vast.aiMarketplace · 99.5% reliableRunPod $0.50/h Rent on Vast.ai
1× RTX 5060 Ti 16GB16 GB~121 tok/sFaster than you read$0.12$0.28Vast.aiMarketplace · 99.1% reliableRent on Vast.ai
1× RTX 3090 Ti24 GB~274 tok/sFaster than you read$0.19$0.19Vast.aiMarketplace · 99.6% reliableRunPod $0.27/h Rent on Vast.ai
2× RTX 407024 GB~274 tok/sFaster than you read$0.19$0.19Vast.aiMarketplace · 99.8% reliableRent on Vast.ai
1× RTX 5070 Ti16 GB~243 tok/sFaster than you read$0.20$0.23Vast.aiMarketplace · 99.9% reliableRent on Vast.ai
2× RTX 307016 GB~243 tok/sFaster than you read$0.22$0.25Vast.aiMarketplace · 99.7% reliableRunPod $0.26/h Rent on Vast.ai

No GPU at all

Use gpt-oss-safeguard-20b by the token

Here the rented card is as cheap as the API or cheaper per token ($0.13 per million), and it keeps your data on a machine you control. The API is still the easier start: nothing to set up and nothing to switch off.

First time renting a GPU? How it works, in four steps
  1. Create an account and add credit. Both providers are prepaid: $10 is enough to try any card on this page for hours. A new RunPod account that arrives through a referral link gets a one-time $5 credit when it first adds $10.
  2. Pick the card and the count shown here, and a template: llama.cpp or vLLM for an OpenAI-compatible API, or one with a web chat if you only want to talk to the model. “How to launch it” below gives the exact image and command.
  3. Wait for the download. The weights are fetched on the machine; tens of gigabytes take a few minutes on a data-centre connection. You pay by the second from the moment it starts.
  4. Stop it when you are done. A running machine bills even when idle. A stopped one costs nothing per hour, though its disk is usually still billed until you delete it.
How to launch it

No Q4_K_M file of gpt-oss-safeguard-20b is catalogued here, so there is no exact command to vouch for. Start the machine from the llama.cpp server image ghcr.io/ggml-org/llama.cpp:server-cuda and point -hf at a Q4_K_M build from the publisher or a quantizer you trust — search Hugging Face for one. Check its file size against the memory figure on this page before you rent.

Every machine holds the whole model at Q4_K_M and 8,192 tokens of context, no offload. Speeds are this site’s single-stream decode estimates; prices are what each provider’s own API quoted, on-demand, for the whole machine. Vast hosts below 98% measured reliability are left out.

Referral links Vast.ai, RunPod and Novita pay us a share of what you spend if you sign up through these buttons. It costs you nothing, and it never decides an order or a recommendation: both are computed from the live price and the speed, and options that pay us nothing are listed and recommended on the same terms. How we rank

Buying a card instead, or paying by the token? Compare a year of gpt-oss-safeguard-20b three ways →

Under the hood

Architecture, read from the publisher’s file

The numbers every figure above is computed from, with the file they came from.

Architecture

✓ Architecture read from the published config.json

Retrieved 2026-09-01 at pinned commit 8a11e17b25c9; the native packed/fused checkpoint metadata is not used as a logical parameter total.

Architecture
gpt_oss
Layers
24
Hidden size
2,880
Attention heads
64
KV heads
8
Head dimension
64
Feed-forward width
2,880
Vocabulary
201,088
Context ceiling
131,072
Sliding window
128
RoPE theta
150,000
Experts
32
Experts / token
4

Where the memory goes

tokenembedding× 24 decoder blocksGrouped-query attention64 query · 8 KV headswindow 128Routed experts4 of 32 per tokenall residentoutputprojectiongrows with contextcapacity ≠ traffic

Grouped-query attention shares each key/value head across 8 query heads, so the KV cache is 13% of what multi-head attention would need at the same context.

Open the exact official config.json ↗

More from OpenAI

gpt-oss-120bgpt-oss-20b

Use this answer in your own product

As JSON, from the API
curl -s https://llmbottleneck.com/v1/analyze \
  -H "Authorization: Bearer $LLMB_KEY" -H "content-type: application/json" \
  -d '{"model":"openai-gpt-oss-safeguard-20b","quantization":"Q4_K_M","context":8192,"hardware":"nvidia-geforce-rtx-4090"}'

Same engine, same evidence, every field sourced. Free key in one step →

On your page, as a widget
<script src="https://llmbottleneck.com/widget.js"
  data-model="openai-gpt-oss-safeguard-20b" data-quantization="Q4_K_M"
  data-hardware="nvidia-geforce-rtx-4090" data-context="8192"></script>

No key needed. Unbranded, with your own buy button, on Pro and Business →

Citing this page

LLM Bottleneck. “gpt-oss-safeguard-20b VRAM and hardware requirements.” Architecture from openai/gpt-oss-safeguard-20b at revision 8a11e17b25c9, retrieved 2026-09-01. https://llmbottleneck.com/models/openai-gpt-oss-safeguard-20b

Every figure above is either the published value or a reconstruction whose measured error is on the accuracy page.

llmbottleneck
catalogue 2026-10-03models 327devices 135