Skip to content
Local AI Wizard
Your machine
Theme

gpt-oss-20b

OpenAI · openai/gpt-oss-20b

Q4 file 11.6 GB (Q4_K_M) · runs well from 24 GB · Weights license: Apache 2.0

LMArena · text1,317 ± 6Rank 269 · release 2026-10-22 · its own scale
Memory needed~16 GBQ4 weights + working room, before the machine's reserve
Parameters21 Bmanufacturer specification · 3.6B active
Context131,072 tokensPublished for this model

Quality

LMArena · text · rating
1,317 ± 6Confidence interval 1,311–1,324 · 10,417 votes · rank 269 of the release · Arena does not state the reasoning effort behind the votesLMArena 2026-10-22 · CC BY 4.0

No result in LiveBench 2026-06-25 Not in LiveBench 2026-06-25

Reported by the publisher

The publisher's card states no benchmark result for this checkpoint (read 2026-10-25). not published on the card · read 2026-10-25

Run it

Run it

File Q4_K_M · 11.6 GB · unsloth/gpt-oss-20b-GGUF · What is Q4?

Measured 112–283 tok/s · MXFP4 · 7 machines · 3 sources · all rows

ollama run hf.co/unsloth/gpt-oss-20b-GGUF:Q4_K_M
  1. Install Ollama
  2. Run the command above
  3. Ask: Explain what a context window is, in two sentences.

Runs on

Fast memory holds it 57

The planning estimate fits the memory the model can use quickly: graphics card VRAM, or unified memory within the macOS GPU share.

Show all 57

Within memory, outside the fast part 47

Within the machine's memory, outside its fast part: on a graphics card the layers that do not fit run on the CPU from PC RAM (the listed RAM, or 32 GB assumed); on a Mac, above the macOS default GPU share. Not a speed figure.

Show all 47

One memory tier short 12

The same machines with the next memory size up would hold it in fast memory.

Show all 12

gpt-oss-20b is a text model from OpenAI with 21B parameters, 3.6B of them active per token (mixture of experts). It was released on August 5, 2025 and reads up to 131,072 tokens of context. Its Q4 weights plus working room come to about 16 GB before the machine's reserve; the smallest reference machines whose fast memory holds that are Mac mini M4 · 24 GB unified (~22 GB with its reserve) and RTX 4090 · 24 GB VRAM (~18 GB with its reserve). We collected 17 downloadable GGUF and MLX files from 2 publisher sources; the most downloaded GGUF source is unsloth's. It has no result in LiveBench 2026-06-25. Its LMArena text rating is 1,317 ± 6 (release 2026-10-22), a separate human-preference scale.

Hardware profiles grouped by the Q4 planning estimate on each machine (dedicated GPUs reserve 2 GB; unified and system memory 25%, 2 to 8 GB): the fast memory holds it, it is within memory but outside the fast part (layers offloaded to PC RAM, or above the macOS GPU share), or the machine is one memory tier short. Each card opens its hardware profile. Not a measured fit or speed.

On the reference machines

gpt-oss-20b on the reference machines
Where Q4 sits
Mac mini M2 · 8 GB unifiedLarger than its memoryNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column~18 GBLarger than its memoryNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column
Mac mini M4 · 16 GB unifiedLarger than its memoryNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column~20 GBLarger than its memoryNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column
Mac mini M4 · 24 GB unifiedFast memory holds itQ8_0GGUF · 12.1 GB · needs ~21 GB131,072 tokens~22 GBFast memory holds itQ8_0GGUF · 12.1 GB · needs ~21 GB131,072 tokens
Mac mini M4 · 32 GB unifiedFast memory holds itQ8_0GGUF · 12.1 GB · needs ~23 GB131,072 tokens~24 GBFast memory holds itQ8_0GGUF · 12.1 GB · needs ~23 GB131,072 tokens
Mac mini M5 Pro · 64 GB unifiedFast memory holds itQ8_0GGUF · 12.1 GB · needs ~23 GB131,072 tokens~24 GBFast memory holds itQ8_0GGUF · 12.1 GB · needs ~23 GB131,072 tokens
Mac Studio M5 Max · 128 GB unifiedFast memory holds itQ8_0GGUF · 12.1 GB · needs ~23 GB131,072 tokens~24 GBFast memory holds itQ8_0GGUF · 12.1 GB · needs ~23 GB131,072 tokens
Mac Studio · M5 Ultra · 36-core CPU / 80-core GPU · 512 GB unifiedFast memory holds itQ8_0GGUF · 12.1 GB · needs ~23 GB131,072 tokens~24 GBFast memory holds itQ8_0GGUF · 12.1 GB · needs ~23 GB131,072 tokens
RTX 4060 · 8 GB VRAMIn memory, outside the fast partNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column~18 GBIn memory, outside the fast partNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column
RTX 3060 · 12 GB VRAMIn memory, outside the fast partNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column~18 GBIn memory, outside the fast partNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column
RTX 5080 · 16 GB VRAMIn memory, outside the fast partNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column~18 GBIn memory, outside the fast partNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column
RTX 4090 · 24 GB VRAMFast memory holds itQ8_0GGUF · 12.1 GB · needs ~17 GB131,072 tokens~18 GBFast memory holds itQ8_0GGUF · 12.1 GB · needs ~17 GB131,072 tokens
RTX 5090 · 32 GB VRAMFast memory holds itQ8_0GGUF · 12.1 GB · needs ~17 GB131,072 tokens~18 GBFast memory holds itQ8_0GGUF · 12.1 GB · needs ~17 GB131,072 tokens
DGX H100 · 8 × 80 GBFast memory holds itQ8_0GGUF · 12.1 GB · needs ~31 GB131,072 tokens~32 GBFast memory holds itQ8_0GGUF · 12.1 GB · needs ~31 GB131,072 tokens

The file column uses the site's quant ladder (Q8_0 to Q3_K_M, MLX 8 to 4-bit on a Mac): the largest collected file whose size × 1.2 plus the context cache and the machine's reserve fits the fast memory. Planning estimates, not measured runs.

Plan another quantization or context

Consider it for

Compare gpt-oss-20b using its publisher card. No overall score appears in the LiveBench 2026-06-25 snapshot.

Check first

OpenAI ships the expert weights in MXFP4 (4-bit) and its card says the model runs within 16 GB of memory. The site's generic Q4 estimate is a separate figure; Runs on adds each machine's reserve to it.

Versions to download

Files you can download for this checkpoint. The section opens on the most downloaded collected GGUF source, unsloth, at Q4_K_M. On a Mac it opens on the most downloaded MLX source. Sizes are the Hub's published bytes. What is Q4?

unsloth

unsloth/gpt-oss-20b-GGUF · gpt-oss-20b-Q4_K_M.gguf · 11.6 GB

Run this file

ollama run hf.co/unsloth/gpt-oss-20b-GGUF:Q4_K_M
llama-server -hf unsloth/gpt-oss-20b-GGUF:Q4_K_M
lms get https://huggingface.co/unsloth/gpt-oss-20b-GGUF
  • mlx-lm: mlx-lm runs MLX repositories; pick the MLX version on a Mac

Quality scores on this page belong to the original checkpoint; these files are not scored or ranked. File sizes read 2026-10-23 · download counts read 2026-10-17 · commands: Hugging Face Hub docs: Use Ollama with any GGUF model · llama.cpp server README (-hf <user>/<model>[:quant]) · LM Studio docs: lms get with a Hugging Face URL · mlx-lm README (mlx_lm.generate --model)

Fine-tunes, adapters and other versions570 fine-tunes
Quantized versions
252
Fine-tunes
570
Adapters · merges
479 · 7

Browse them on the Hugging Face model tree · read 2026-10-23

Uncensored and abliterated versions8

Some community uploads remove the model's safety refusals. Quality may drop, the original publisher does not support them, claims about them are the uploader's, and the original license still applies. This site never recommends or ranks them.

8 derivative repositories have “uncensored” or “abliterated” in their name (read 2026-10-23).

Measured speed

Tokens per second by machine

Output tok/s · 9 machines · fastest 5 of 20 · every row in the table · What is a token?

RTX 5090 · Q4_K_M, 393.3 ± 2.62 · Q4_K_M · llama.cpp CUDA · llama.cpp commit 3e4dfa8f3 (b11658)393.3
RTX 5090 · Q8_0, 374.72 ± 2.73 · Q8_0 · llama.cpp CUDA · llama.cpp commit 3e4dfa8f3 (b11658)374.7
RTX 5090 · MXFP4, 282.51 ± 0.44 · MXFP4 (native gpt-oss weights) · llama.cpp CUDA · llama.cpp commit a094f38 (6210)282.5
RTX 4090 24 GB · Q4_K_M, 277.95 ± 1.50 · Q4_K_M · llama.cpp CUDA · llama.cpp commit 3e4dfa8f3 (b11658)278.0
RTX 4090 24 GB · Q8_0, 258.4 ± 1.92 · Q8_0 · llama.cpp CUDA · llama.cpp commit 3e4dfa8f3 (b11658)258.4
Measured speed
RTX 5090NVIDIA GeForce RTX 5090 · 32 GB (the card's only memory size; rented host)393.3 ± 2.62Q4_K_M0 / 128 · Single llama-bench processllama.cpp CUDAllama.cpp commit 3e4dfa8f3 (b11658)Flash Attention enabledunsloth/gpt-oss-20b-GGUF/gpt-oss-20b-Q4_K_M.ggufData: Local LLM Advisor (localllmadvisor.com) · CC BY 4.0 · 2026-10-25
RTX 5090NVIDIA GeForce RTX 5090 · 32 GB (the card's only memory size; rented host)374.72 ± 2.73Q8_00 / 128 · Single llama-bench processllama.cpp CUDAllama.cpp commit 3e4dfa8f3 (b11658)Flash Attention enabledunsloth/gpt-oss-20b-GGUF/gpt-oss-20b-Q8_0.ggufData: Local LLM Advisor (localllmadvisor.com) · CC BY 4.0 · 2026-10-25
RTX 5090NVIDIA GeForce RTX 5090 · 32 GB282.51 ± 0.44MXFP4 (native gpt-oss weights)0 / 128 · Single llama-bench process; server concurrency not measuredllama.cpp CUDAllama.cpp commit a094f38 (6210)Flash Attention enabledgpt-oss-20b-mxfp4.gguf (ggml-org/gpt-oss-20b-GGUF, the checkpoint the guide links); llama-bench model “gpt-oss 20B MXFP4 MoE”, 11.27 GiBllama.cpp gpt-oss guide · maintainer result · 2026-10-20
RTX 4090 24 GBNVIDIA GeForce RTX 4090 · 24 GB (the card's only memory size; rented host)277.95 ± 1.50Q4_K_M0 / 128 · Single llama-bench processllama.cpp CUDAllama.cpp commit 3e4dfa8f3 (b11658)Flash Attention enabledunsloth/gpt-oss-20b-GGUF/gpt-oss-20b-Q4_K_M.ggufData: Local LLM Advisor (localllmadvisor.com) · CC BY 4.0 · 2026-10-25
RTX 4090 24 GBNVIDIA GeForce RTX 4090 · 24 GB (the card's only memory size; rented host)258.4 ± 1.92Q8_00 / 128 · Single llama-bench processllama.cpp CUDAllama.cpp commit 3e4dfa8f3 (b11658)Flash Attention enabledunsloth/gpt-oss-20b-GGUF/gpt-oss-20b-Q8_0.ggufData: Local LLM Advisor (localllmadvisor.com) · CC BY 4.0 · 2026-10-25
RTX 5080 16 GBNVIDIA GeForce RTX 5080 · 16 GB (the card's only memory size; rented host)253.49 ± 1.35Q4_K_M0 / 128 · Single llama-bench processllama.cpp CUDAllama.cpp commit 3e4dfa8f3 (b11658)Flash Attention enabledunsloth/gpt-oss-20b-GGUF/gpt-oss-20b-Q4_K_M.ggufData: Local LLM Advisor (localllmadvisor.com) · CC BY 4.0 · 2026-10-25
RTX 5080 16 GBNVIDIA GeForce RTX 5080 · 16 GB (the card's only memory size; rented host)238.83 ± 1.12Q8_00 / 128 · Single llama-bench processllama.cpp CUDAllama.cpp commit 3e4dfa8f3 (b11658)Flash Attention enabledunsloth/gpt-oss-20b-GGUF/gpt-oss-20b-Q8_0.ggufData: Local LLM Advisor (localllmadvisor.com) · CC BY 4.0 · 2026-10-25
RTX 5070 Ti 16 GBNVIDIA GeForce RTX 5070 Ti · 16 GB (the card's only memory size; rented host)236.89 ± 2.18Q4_K_M0 / 128 · Single llama-bench processllama.cpp CUDAllama.cpp commit 3e4dfa8f3 (b11658)Flash Attention enabledunsloth/gpt-oss-20b-GGUF/gpt-oss-20b-Q4_K_M.ggufData: Local LLM Advisor (localllmadvisor.com) · CC BY 4.0 · 2026-10-25
RTX 5070 Ti 16 GBNVIDIA GeForce RTX 5070 Ti · 16 GB (the card's only memory size; rented host)223.82 ± 2.47Q8_00 / 128 · Single llama-bench processllama.cpp CUDAllama.cpp commit 3e4dfa8f3 (b11658)Flash Attention enabledunsloth/gpt-oss-20b-GGUF/gpt-oss-20b-Q8_0.ggufData: Local LLM Advisor (localllmadvisor.com) · CC BY 4.0 · 2026-10-25
RTX 4090 24 GBNVIDIA GeForce RTX 4090 · 24 GB221.95 ± 6.34MXFP4 (native gpt-oss weights)0 / 128 · Single llama-bench process; server concurrency not measuredllama.cpp CUDAllama.cpp commit a094f38 (6210)Flash Attention enabledgpt-oss-20b-mxfp4.gguf (ggml-org/gpt-oss-20b-GGUF, the checkpoint the guide links); llama-bench model “gpt-oss 20B MXFP4 MoE”, 11.27 GiBllama.cpp gpt-oss guide · maintainer result · 2026-10-20
RTX 3090 24 GBNVIDIA GeForce RTX 3090 · 24 GB (the card's only memory size; rented host)214.34 ± 1.51Q4_K_M0 / 128 · Single llama-bench processllama.cpp CUDAllama.cpp commit 3e4dfa8f3 (b11658)Flash Attention enabledunsloth/gpt-oss-20b-GGUF/gpt-oss-20b-Q4_K_M.ggufData: Local LLM Advisor (localllmadvisor.com) · CC BY 4.0 · 2026-10-25
RTX 5080 16 GBNVIDIA GeForce RTX 5080 · 16 GB204.85 ± 0.23MXFP4 (native gpt-oss weights)0 / 128 · Single llama-bench process; server concurrency not measuredllama.cpp CUDAllama.cpp commit a094f38 (6210)Flash Attention enabledgpt-oss-20b-mxfp4.gguf (ggml-org/gpt-oss-20b-GGUF, the checkpoint the guide links); llama-bench model “gpt-oss 20B MXFP4 MoE”, 11.27 GiBllama.cpp gpt-oss guide · maintainer result · 2026-10-20
RTX 3090 24 GBNVIDIA GeForce RTX 3090 · 24 GB (the card's only memory size; rented host)204.22 ± 1.07Q8_00 / 128 · Single llama-bench processllama.cpp CUDAllama.cpp commit 3e4dfa8f3 (b11658)Flash Attention enabledunsloth/gpt-oss-20b-GGUF/gpt-oss-20b-Q8_0.ggufData: Local LLM Advisor (localllmadvisor.com) · CC BY 4.0 · 2026-10-25
RTX 5070 Ti 16 GBNVIDIA GeForce RTX 5070 Ti · 16 GB189.45 ± 0.09MXFP4 (native gpt-oss weights)0 / 128 · Single llama-bench process; server concurrency not measuredllama.cpp CUDAllama.cpp commit a094f38 (6210)Flash Attention enabledgpt-oss-20b-mxfp4.gguf (ggml-org/gpt-oss-20b-GGUF, the checkpoint the guide links); llama-bench model “gpt-oss 20B MXFP4 MoE”, 11.27 GiBllama.cpp gpt-oss guide · maintainer result · 2026-10-20
RTX 4080 SUPER 16 GBNVIDIA GeForce RTX 4080 SUPER · 16 GB186.51 ± 0.33MXFP4 (native gpt-oss weights)0 / 128 · Single llama-bench process; server concurrency not measuredllama.cpp CUDAllama.cpp commit 009b709 (6316)Flash Attention enabledgpt-oss-20b-mxfp4.gguf (ggml-org/gpt-oss-20b-GGUF, the checkpoint the guide links); llama-bench model “gpt-oss 20B MXFP4 MoE”, 11.27 GiBllama.cpp gpt-oss guide · maintainer result · 2026-10-20
RTX 3090 24 GBNVIDIA GeForce RTX 3090 · 24 GB161.77 ± 0.56MXFP4 (native gpt-oss weights)0 / 128 · Single llama-bench process; server concurrency not measuredllama.cpp CUDAllama.cpp commit a094f38 (6210)Flash Attention enabledgpt-oss-20b-mxfp4.gguf (ggml-org/gpt-oss-20b-GGUF, the checkpoint the guide links); llama-bench model “gpt-oss 20B MXFP4 MoE”, 11.27 GiBllama.cpp gpt-oss guide · maintainer result · 2026-10-20
RTX 5060 Ti 16 GBNVIDIA GeForce RTX 5060 Ti · 16 GB111.51 ± 0.05MXFP4 (native gpt-oss weights)0 / 128 · Single llama-bench process; server concurrency not measuredllama.cpp CUDAllama.cpp commit 9ef6b0b (6208)Flash Attention enabledgpt-oss-20b-mxfp4.gguf (ggml-org/gpt-oss-20b-GGUF, the checkpoint the guide links); llama-bench model “gpt-oss 20B MXFP4 MoE”, 11.27 GiBllama.cpp gpt-oss guide · maintainer result · 2026-10-20
RTX 4060 Ti 16GBNVIDIA GeForce RTX 4060 Ti · 16 GB (the dataset id names 16 GB; rented host)103.72 ± 0.27Q4_K_M0 / 128 · Single llama-bench processllama.cpp CUDAllama.cpp commit 3e4dfa8f3 (b11658)Flash Attention enabledunsloth/gpt-oss-20b-GGUF/gpt-oss-20b-Q4_K_M.ggufData: Local LLM Advisor (localllmadvisor.com) · CC BY 4.0 · 2026-10-25
RTX 4060 Ti 16GBNVIDIA GeForce RTX 4060 Ti · 16 GB (the dataset id names 16 GB; rented host)94.43 ± 0.13Q8_00 / 128 · Single llama-bench processllama.cpp CUDAllama.cpp commit 3e4dfa8f3 (b11658)Flash Attention enabledunsloth/gpt-oss-20b-GGUF/gpt-oss-20b-Q8_0.ggufData: Local LLM Advisor (localllmadvisor.com) · CC BY 4.0 · 2026-10-25
DGX SparkDGX Spark · 1× NVIDIA GB1083.43 ± 0.59MXFP4 MoE0 / 32 · Not reported by llama-benchCUDAllama.cpp 11fb327bf (7941); CUDAFlash Attention enabledggml-org/gpt-oss-20b-GGUFllama.cpp bench · 2026-09-25

Sources: llama.cpp bench · llama.cpp gpt-oss guide · maintainer result · Data: Local LLM Advisor (localllmadvisor.com) · CC BY 4.0 · not estimated

How to run

The memory needed above is the Q4 load before a machine's reserve. Formula: parameters × 0.625 bytes, plus 20% working allowance; each machine then adds its system reserve: 25% of unified or system memory, 2 to 8 GB, or 2 GB per GPU on dedicated cards. This is neither measured VRAM nor checkpoint download size.

Q4 planning estimate only. Check checkpoint format, runtime support, KV cache, context, batch size and offload before choosing hardware.

Evidence level: Generic estimate only; no execution reproduced by this site.

Publisher sampling defaults
— generation_config.json states no sampling values for this checkpoint
Chat template markers
tool calls · thinkingmarkers found in the template, not a test of the feature · read 2026-10-23source
Before downloading
  1. Check the exact checkpoint and license linked here.
  2. Confirm a compatible runtime and precision or quantization for your platform.
  3. Allow for context, batch size and any extra media components beyond the weights.

Technical details

What the source and the catalog store for this checkpoint. Missing fields are left out, never zero.

Checkpoint / version
openai/gpt-oss-20b
Task and modality
language and reasoning · text
Count and basis
21B · manufacturer specification · 3.6B activeopenai/gpt-oss-20b model card (Hugging Face) · read 2026-10-17source
Context
131,072 tokens published for this modelgpt-oss-120b & gpt-oss-20b Model Card (OpenAI, PDF) · read 2026-10-09source
Architecture
MoEopenai/gpt-oss-20b model card (Hugging Face) · read 2026-10-17source
Released
gpt-oss-120b & gpt-oss-20b Model Card (OpenAI, PDF) · read 2026-10-09source
Runtime and precision
No local execution verified.
Commercial use
Allowed Apache License 2.0 text (SPDX) · read 2026-10-18license text
Every stored field12 fields
Revision checked
6cee5e81ee83917806bbde320786a8fb61efebee
Architecture
GptOssForCausalLM
Library
transformers
Hub task tag
text-generation
Distributions in this group
mlx-community/gpt-oss-20b-MXFP4-Q8
Formats
safetensors
Access
Open download
Publisher kind
Official
Repository created
2025-08-04creation is not the model's release
Repository updated
2025-08-26a commit is not a release date
Hub downloads (30 days) · likes
5,766,151 · 5,218this repository only · read 2026-10-25
Catalog state
Hub config or weights found; no inventory benchmark link

Sources

Identity and published specifications come from the checkpoint page. Benchmark scores belong to the named release and do not measure speed on your computer.

Catalog reviewed: 2026-10-17. Benchmark release: 2026-06-25; snapshot collected: 2026-10-18.