What's the best quantization for GPT OSS 20B MXFP4 Q8?

For GPT OSS 20B MXFP4 Q8, Q4_K_M (12.4 GB) offers the best balance of quality and VRAM usage. Q4_K_L (12.6 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 5.9 GB.

Can I run GPT OSS 20B MXFP4 Q8 on a Mac?

GPT OSS 20B MXFP4 Q8 requires at least 5.9 GB at IQ2_XXS, which exceeds the unified memory of most consumer Macs. You would need a Mac Studio or Mac Pro with a high-memory configuration.

Can I run GPT OSS 20B MXFP4 Q8 locally?

Yes — GPT OSS 20B MXFP4 Q8 can run locally on consumer hardware. At Q4_K_M quantization it needs 12.4 GB of VRAM. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is GPT OSS 20B MXFP4 Q8?

At Q4_K_M, GPT OSS 20B MXFP4 Q8 can reach ~236 tok/s on AMD Instinct MI300X. On NVIDIA GeForce RTX 4090: ~53 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

What's the download size of GPT OSS 20B MXFP4 Q8?

At Q4_K_M, the download is about 12.00 GB. The full-precision Q8_0 version is 20.00 GB. The smallest option (IQ2_XXS) is 5.50 GB.

MLX Community·GPT-OSS·GptOssForCausalLM

GPT OSS 20B MXFP4 Q8 — Hardware Requirements & GPU Compatibility

Chat

An MLX-optimized MXFP4-Q8 quantized version of OpenAI's GPT-OSS 20B, converted by MLX Community for Apple Silicon Macs. This model uses a mixed-precision quantization scheme with MXFP4 weights and Q8 attention, designed to maximize performance on Apple's unified memory architecture while keeping the memory footprint manageable. GPT-OSS 20B is OpenAI's open-source entry at 20 billion parameters, and this MLX conversion makes it straightforward to run natively on M-series Macs without any CUDA dependency. Users with 32GB or more of unified memory should be able to run this model comfortably for general-purpose chat, writing, and reasoning tasks.

542.2K downloads 36 likesAug 2025131K context

Based on GPT OSS 20B

Specifications

Publisher: MLX Community
Family: GPT-OSS
Parameters: 20B
Architecture: GptOssForCausalLM
Context Length: 131,072 tokens
Vocabulary Size: 201,088
Release Date: 2025-08-29
License: Apache 2.0

Get Started

HuggingFace

mlx-community/gpt-oss-20b-MXFP4-Q8

How Much VRAM Does GPT OSS 20B MXFP4 Q8 Need?

Select a quantization to see compatible GPUs below.

Quantization	Bits	VRAM	+ Context	File Size	Quality
IQ2_XXS	2.20	5.9 GB	10.3 GB	5.50 GB	Importance-weighted 2-bit, extreme compression — significant quality loss
IQ2_XS	2.40	6.4 GB	10.8 GB	6.00 GB	Importance-weighted 2-bit, extra small
IQ2_S	2.50	6.6 GB	11.1 GB	6.25 GB	Importance-weighted 2-bit, small
IQ2_M	2.70	7.1 GB	11.6 GB	6.75 GB	Importance-weighted 2-bit, medium
IQ3_XXS	3.10	8.1 GB	12.6 GB	7.75 GB	Importance-weighted 3-bit
IQ3_XS	3.30	8.6 GB	13.1 GB	8.25 GB	Importance-weighted 3-bit, extra small
Q2_K	3.40	8.9 GB	13.3 GB	8.50 GB	2-bit quantization with K-quant improvements
Q3_K_S	3.50	9.1 GB	13.6 GB	8.75 GB	3-bit small quantization
IQ3_M	3.60	9.4 GB	13.8 GB	9.00 GB	Importance-weighted 3-bit, medium
Q3_K_M	3.90	10.1 GB	14.6 GB	9.75 GB	3-bit medium quantization
Q4_0	4.00	10.4 GB	14.8 GB	10.00 GB	4-bit legacy quantization
Q3_K_L	4.10	10.6 GB	15.1 GB	10.25 GB	3-bit large quantization
IQ4_XS	4.30	11.1 GB	15.6 GB	10.75 GB	Importance-weighted 4-bit, compact
Q4_1	4.50	11.6 GB	16.1 GB	11.25 GB	4-bit legacy quantization with offset
Q4_K_S	4.50	11.6 GB	16.1 GB	11.25 GB	4-bit small quantization
IQ4_NL	4.50	11.6 GB	16.1 GB	11.25 GB	Importance-weighted 4-bit, non-linear
Q4_K_M	4.80	12.4 GB	16.8 GB	12.00 GB	4-bit medium quantization — most popular sweet spot
Q4_K_L	4.90	12.6 GB	17.1 GB	12.25 GB	4-bit large quantization
Q5_K_S	5.50	14.1 GB	18.6 GB	13.75 GB	5-bit small quantization
Q5_K_M	5.70	14.6 GB	19.1 GB	14.25 GB	5-bit medium quantization — good quality/size tradeoff
Q5_K_L	5.80	14.9 GB	19.3 GB	14.50 GB	5-bit large quantization
Q6_K	6.60	16.9 GB	21.3 GB	16.50 GB	6-bit quantization, very good quality
Q8_0	8.00	20.4 GB	24.8 GB	20.00 GB	8-bit quantization, near-lossless

Which GPUs Can Run GPT OSS 20B MXFP4 Q8?

Q4_K_M · 12.4 GB

Show professional

GPT OSS 20B MXFP4 Q8 (Q4_K_M) requires 12.4 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 17+ GB is recommended. Using the full 131K context window can add up to 4.5 GB, bringing total usage to 16.8 GB. 17 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 5080.

Runs great

— Plenty of headroom

NVIDIA GeForce RTX 5090~94 tok/s NVIDIA GeForce RTX 3090 Ti~53 tok/s NVIDIA GeForce RTX 4090~53 tok/s NVIDIA GeForce RTX 3090~49 tok/s AMD Radeon RX 7900 XTX~43 tok/s AMD Radeon RX 7900 XT~36 tok/s

Decent

— Enough VRAM, may be tight

NVIDIA GeForce RTX 5080~50 tok/s NVIDIA GeForce RTX 5070 Ti~47 tok/s NVIDIA GeForce RTX 4080 SUPER~39 tok/s NVIDIA GeForce RTX 4080~38 tok/s NVIDIA GeForce RTX 4070 Ti SUPER~35 tok/s AMD Radeon RX 7800 XT~28 tok/s AMD Radeon RX 6800~23 tok/s AMD Radeon RX 6800 XT~23 tok/s AMD Radeon RX 6900 XT~23 tok/s Intel Arc A770 16GB~23 tok/s NVIDIA GeForce RTX 4060 Ti 16GB~15 tok/s

Which Devices Can Run GPT OSS 20B MXFP4 Q8?

Q4_K_M · 12.4 GB

27 devices with unified memory can run GPT OSS 20B MXFP4 Q8, including NVIDIA DGX H100, NVIDIA DGX A100 640GB, Mac Mini M4 (16 GB).

Runs great

— Plenty of headroom

Decent

— Enough memory, may be tight

Mac Mini M4 (16 GB)~6 tok/s MacBook Air 13" M4 (16 GB)~6 tok/s MacBook Air 15" M4 (16 GB)~6 tok/s MacBook Pro 14" M4 (16 GB)~6 tok/s iPad Pro M4 13" (16 GB)~6 tok/s MacBook Air 13" M3 (16 GB)~5 tok/s

Related Models

Frequently Asked Questions

How much VRAM does GPT OSS 20B MXFP4 Q8 need?: GPT OSS 20B MXFP4 Q8 requires 12.4 GB of VRAM at Q4_K_M, or 20.4 GB at Q8_0. Full 131K context adds up to 4.5 GB (16.8 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 20B × 4.8 bits ÷ 8 = 12 GB
KV Cache + Overhead ≈ 0.4 GB (at 2K context + ~0.3 GB framework)
KV Cache + Overhead ≈ 4.8 GB (at full 131K context)
VRAM usage by quantization
Q4_K_M
12.4 GB
Q4_K_M + full context
16.8 GB
Learn more about VRAM estimation →
What's the best quantization for GPT OSS 20B MXFP4 Q8?: For GPT OSS 20B MXFP4 Q8, Q4_K_M (12.4 GB) offers the best balance of quality and VRAM usage. Q4_K_L (12.6 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 5.9 GB.
VRAM requirement by quantization
IQ2_XXS
5.9 GB~53%
Q2_K
8.9 GB~75%
Q3_K_L
10.6 GB~86%
Q4_K_M ★
12.4 GB~89%
Q4_K_L
12.6 GB~90%
Q8_0
20.4 GB~99%
★ Recommended — best balance of quality and VRAM usage.
Learn more about quantization →
Can I run GPT OSS 20B MXFP4 Q8 on a Mac?: GPT OSS 20B MXFP4 Q8 requires at least 5.9 GB at IQ2_XXS, which exceeds the unified memory of most consumer Macs. You would need a Mac Studio or Mac Pro with a high-memory configuration.
Can I run GPT OSS 20B MXFP4 Q8 locally?: Yes — GPT OSS 20B MXFP4 Q8 can run locally on consumer hardware. At Q4_K_M quantization it needs 12.4 GB of VRAM. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is GPT OSS 20B MXFP4 Q8?: At Q4_K_M, GPT OSS 20B MXFP4 Q8 can reach ~236 tok/s on AMD Instinct MI300X. On NVIDIA GeForce RTX 4090: ~53 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: AMD Instinct MI300X → 5300 ÷ 12.4 × 0.55 = ~236 tok/s
Estimated speed at Q4_K_M (12.4 GB)
AMD Instinct MI300X
~236 tok/s
NVIDIA GeForce RTX 4090
~53 tok/s
NVIDIA H100 SXM
~176 tok/s
AMD Instinct MI250X
~146 tok/s
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
Learn more about tok/s estimation →
What's the download size of GPT OSS 20B MXFP4 Q8?: At Q4_K_M, the download is about 12.00 GB. The full-precision Q8_0 version is 20.00 GB. The smallest option (IQ2_XXS) is 5.50 GB.