MicrosoftQwen3VLForConditionalGeneration

AesCode 8B — Hardware Requirements & GPU Compatibility

VisionCodeFunctions

AesCode-8B is the smaller of Microsoft's two AesCode models, which generate slides, posters, dashboards, and infographics as structured, editable HTML and CSS. It starts from Qwen3-VL-8B-Instruct and is trained with supervised fine-tuning and then reinforcement learning across seven reward channels. It can use an image generated from the same prompt as an aesthetic reference while taking the actual content from the text. On the paper's 300-sample infographic test, it scored 82.94 overall, a large gain over the 51.93 of its reference-conditioned base model. At about 8.8 billion parameters, it fits an 8 GB GPU at Q4 and runs well on 12 GB or more. The context window is 262,144 tokens, and the license is Apache 2.0. It accepts images and text. This is a design-to-code specialist rather than a general assistant, suited to producing editable visual documents locally. The larger AesCode-32B scores higher, reaching 85.89 overall.

8.8B params · dense262K contextApache 2.0Sep 2026587 downloads69 likes10.1K quant downloads

Based on Qwen3 VL 8B Instruct

Your hardware

Verdicts below are for the reference RTX 4090. Set yours to personalise the page.

Detection reads your browser's GPU name with no data sent anywhere.

Runs great on a 24 GB GPU

Runs from 8 GB (Q4_K_M · 16K)Best on 24 GB (Q4_K_M · ~112 tok/s · 64K)Macs from 32 GB (~14 tok/s)

Apache 2.0

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run AesCode 8B

One recommended way. Every other method, OS and quant is in the step-by-step guide.

Other ways

each opens step-by-step

Run in cloud

Fits on RTX 3060 12GB (4 GB headroom) · Q4_K_M

Generation speed: ~40 tok/s; generation speed
Cost per 1M output tokens: $0.48; per 1M output tokens

Deployon Vast.aiBest optionfrom $0.069/hr

Compare GPUs →

Run AesCode 8B

8.0 GB VRAM at 16K context · fits the RTX 4090 (16.0 GB free)

Works on macOS, Windows and Linux.

Install Ollama (Linux)
$curl -fsSL https://ollama.com/install.sh | sh
Install Ollama (macOS / Windows)
Download and run the app; it installs the ollama command.
Download Ollama ↗
Run Q4_K_M
$ollama run hf.co/bartowski/AesCode-8B-GGUF:Q4_K_M
Raise the context window
Type it inside the chat. For the API server instead, start it with OLLAMA_CONTEXT_LENGTH=16384 ollama serve.
$/set parameter num_ctx 16384
A larger context needs more memory than the weights alone.
Use the API
Ollama serves its API on http://localhost:11434.

Works on macOS, Windows and Linux.

Install LM Studio
Desktop app for macOS (Apple silicon), Windows and Linux.
lmstudio.ai ↗
Download the model
In the app press Ctrl/Cmd + Shift + M and paste https://huggingface.co/bartowski/AesCode-8B-GGUF, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/bartowski/AesCode-8B-GGUF@Q4_K_M
Open on Hugging Face ↗
Load and chat
Open the model loader, select the model and set Context Length to 16384. Leave GPU offload at maximum if it fits.
Start the local server (optional)
$lms server start

Works on macOS, Windows and Linux.

Install llama.cpp (macOS)
MacPorts, Nix and conda-forge also work.
$brew install llama.cpp
Install options ↗
Install llama.cpp (Windows)
$winget install llama.cpp
Install llama.cpp (Linux)
Download the prebuilt llama-*-bin-ubuntu-* archive for your GPU backend (CPU, Vulkan) from the releases page, unpack it and run the binaries from that folder. Homebrew (brew install llama.cpp) and conda-forge also work on Linux.
llama.cpp releases ↗
Start the server
Downloads the model on first run. The chat UI and an OpenAI-compatible API are at http://localhost:8080.
$llama-server -hf bartowski/AesCode-8B-GGUF:Q4_K_M -c 16384 -ngl 99 --port 8080

Works on macOS, Windows and Linux.

Option A: Docker Model Runner
Docker Desktop: Settings > AI > Enable Docker Model Runner. Docker Engine on Linux: install the docker-model-plugin package.
$docker model run hf.co/bartowski/AesCode-8B-GGUF:Q4_K_M
Model Runner docs ↗
The quant tag after the repo is how Hugging Face cards write it; if your Docker version rejects it, drop the tag and pick the quant by repo.
Option B: llama.cpp server image (NVIDIA)
Needs the NVIDIA Container Toolkit. Models are cached in the llama-cache volume. Open http://localhost:8080.
$docker run --gpus all -p 8080:8080 -v llama-cache:/root/.cache ghcr.io/ggml-org/llama.cpp:server-cuda -hf bartowski/AesCode-8B-GGUF:Q4_K_M -c 16384 -ngl 99 --host 0.0.0.0 --port 8080
Image tags (cuda13, vulkan, rocm) ↗
Cache path inside the image is not in the docs; remove the -v flag if the volume complains.

Linux only.

Install vLLM
Linux only; on Windows run it inside WSL2. NVIDIA GPU shown; AMD ROCm, Intel and CPU builds have their own install pages.
$uv venv --python 3.12 --seed
$source .venv/bin/activate
$uv pip install vllm --torch-backend=auto
vLLM installation ↗
Serve the model
OpenAI-compatible API at http://localhost:8000/v1.
$vllm serve microsoft/AesCode-8B --max-model-len 16384
Runs the original ~19 GB weights, not the Q4_K_M GGUF.

macOS only.

MLX supports only macOS, so follow these steps for that setup.

Install mlx-lm
$pip install mlx-lm
Chat in the terminal
$mlx_lm.chat --model prithivMLmods/microsoft-AesCode-8B-MLX
Downloads the model on first run. The MLX repo has its own quantisation; it may not match the quant selected above.
Or start a local server
Listens on http://localhost:8080.
$mlx_lm.server --model prithivMLmods/microsoft-AesCode-8B-MLX

Works on macOS, Windows and Linux.

Install Jan
jan.ai ↗
Find the model
Open the Hub tab, paste this model ID in Search, and choose Q4_K_M if offered.
$bartowski/AesCode-8B-GGUF
Use the local API (optional)
Settings > Local API Server > Start Server. It listens on http://127.0.0.1:1337.

Rent GPUs hourly to run Q4_K_M right there. Pricing comes from partner marketplaces and varies.

Available GPU rentals for Q4_K_M, prioritized by value
GPU	Speed	$/1M tokens	Rent
RTX 3060 12GBCheapestCheapest4 GB free	~40 tok/s	$0.48	Deploy on Vast.aifrom $0.069/hr (opens in new tab)	4 GB free · ~40 tok/s · $0.48 / 1M tokens
RTX A4000 (16 GB)8 GB free	~50 tok/s	$0.61	Deploy on Vast.aifrom $0.11/hr (opens in new tab)	16 GB · 8 GB free · ~50 tok/s · $0.61 / 1M tokens
RTX 5060 Ti 16GB8 GB free	~50 tok/s	$0.77	Deploy on Vast.aifrom $0.14/hr (opens in new tab)	8 GB free · ~50 tok/s · $0.77 / 1M tokens
RTX 4060 Ti 16GB8 GB free	~32 tok/s	$1.34	Deploy on Vast.aifrom $0.15/hr (opens in new tab)	8 GB free · ~32 tok/s · $1.34 / 1M tokens
RTX A5000 (24 GB)16 GB free	~85 tok/s	$0.52	Deploy on RunPodfrom $0.16/hr (opens in new tab)	24 GB · 16 GB free · ~85 tok/s · $0.52 / 1M tokens
RTX 3080 Ti (12 GB)4 GB free	~101 tok/s	$0.49	Deploy on RunPodfrom $0.18/hr (opens in new tab)	12 GB · 4 GB free · ~101 tok/s · $0.49 / 1M tokens
Tesla V100 PCIe 16GB8 GB free	~100 tok/s	$0.53	Deploy on RunPodfrom $0.19/hr (opens in new tab)	8 GB free · ~100 tok/s · $0.53 / 1M tokens
RTX 4000 Ada (20 GB)12 GB free	~40 tok/s	$1.39	Deploy on RunPodfrom $0.20/hr (opens in new tab)	20 GB · 12 GB free · ~40 tok/s · $1.39 / 1M tokens

Compare GPUs →

How Much VRAM Does AesCode 8B Need?

Select a quantization level; the listed hardware adheres to it.

VRAM	+32K	+128K	File	Quality	On RTX 4090 (reference)
3.0 GB	7.5 GB	22 GB	2.5 GB	2-bit · i-quant extreme compression — Importance-weighted 2-bit, extreme compression — significant quality loss	Fits: 128K context · ~218 tok/s
3.2 GB	7.8 GB	22.3 GB	2.7 GB	2-bit · i-quant extra small — Importance-weighted 2-bit, extra small	Fits: 128K context · ~203 tok/s
3.3 GB	7.9 GB	22.4 GB	2.9 GB	2-bit · i-quant small — Importance-weighted 2-bit, small	Fits: 128K context · ~196 tok/s
3.6 GB	8.1 GB	22.6 GB	3.1 GB	2-bit · i-quant medium — Importance-weighted 2-bit, medium	Fits: 128K context · ~184 tok/s
4 GB	8.5 GB	23 GB	3.7 GB	3-bit · i-quant — Importance-weighted 3-bit	Fits: 128K context · ~164 tok/s
4.1 GB	8.6 GB	23.1 GB	3.1 GB	2-bit · small K-quant — 2-bit small K-quant	Fits: 128K context · ~159 tok/s
4.2 GB	8.8 GB	23.3 GB	3.9 GB	3-bit · i-quant extra small — Importance-weighted 3-bit, extra small	Fits: 128K context · ~155 tok/s
4.3 GB	8.9 GB	23.4 GB	3.8 GB	3-bit · i-quant small — Importance-weighted 3-bit, small	Fits: 128K context · ~151 tok/s
4.3 GB	8.9 GB	23.4 GB	3.3 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Fits: 128K context · ~151 tok/s
4.4 GB	9 GB	23.5 GB	3.9 GB	3-bit small — 3-bit small quantization	Fits: 128K context · ~148 tok/s
4.5 GB	9.1 GB	23.6 GB	4.5 GB	3-bit · i-quant medium — Importance-weighted 3-bit, medium	Fits: 128K context · ~144 tok/s
4.9 GB	9.4 GB	23.9 GB	4 GB	3-bit medium — 3-bit medium quantization	Fits: 128K context · ~134 tok/s
5.0 GB	9.5 GB	24 GB	4.9 GB	4-bit · legacy — 4-bit legacy quantization	Fits: 64K context · ~131 tok/s
5.1 GB	9.6 GB	24.1 GB	4.2 GB	3-bit large — 3-bit large quantization	Fits: 64K context · ~129 tok/s
5.3 GB	9.8 GB	24.3 GB	4.7 GB	4-bit · i-quant compact — Importance-weighted 4-bit, compact	Fits: 64K context · ~123 tok/s
5.5 GB	10.1 GB	24.6 GB	4.9 GB	4-bit small — 4-bit small quantization	Fits: 64K context · ~119 tok/s
5.5 GB	10.1 GB	24.6 GB	5.2 GB	4-bit · i-quant non-linear — Importance-weighted 4-bit, non-linear	Fits: 64K context · ~119 tok/s
5.5 GB	10.1 GB	24.6 GB	5.3 GB	4-bit · legacy+ — 4-bit legacy quantization with offset	Fits: 64K context · ~119 tok/s
5.9 GB	10.4 GB	24.9 GB	5.2 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Fits: 64K context · ~112 tok/s
6.0 GB	10.5 GB	25 GB	5.6 GB	4-bit large — 4-bit large quantization	Fits: 64K context · ~110 tok/s
6.6 GB	11.2 GB	25.7 GB	5.9 GB	5-bit small — 5-bit small quantization	Fits: 64K context · ~99 tok/s
6.8 GB	11.4 GB	25.9 GB	6.2 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Fits: 64K context · ~96 tok/s
7.8 GB	12.4 GB	26.9 GB	7.1 GB	6-bit, very good — 6-bit quantization, very good quality	Fits: 64K context · ~84 tok/s
9.4 GB	13.9 GB	28.4 GB	8.7 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Fits: 64K context · ~70 tok/s
18.1 GB	22.7 GB	37.2 GB	16.4 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Fits: 32K context · ~36 tok/s

Selected Q4_K_M

4-bit medium quantization

8.0GBat 16K context

Weights: 5.3 GB
KV cache: 2.4 GB
Overhead: 0.3 GB

Fits RTX 4090 (reference) · 16.0 GB free · ~112 tok/s

≥50 tok/s needs ≥450 GB/s · RTX 5090 ~199, RTX 3090 Ti ~112

Which Hardware Can Run AesCode 8B?

Q4_K_M · 5.9 GB

AesCode 8B (Q4_K_M) requires 5.9 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 8+ GB is recommended. Using the full 262K context window can add up to 38.4 GB, bringing total usage to 44.2 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 3070 Ti. 159 devices with unified memory can run AesCode 8B, including NVIDIA DGX H100, NVIDIA DGX A100 640GB, NVIDIA Jetson Orin Nano 8GB (Super).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Typical hardwareQuantVRAMtok/sContext that fits

8 GB · RTX 4060Q4_K_M · 8 GB · ~30 tok/s · 16K8~3016K8 GB tier →

12 GB · RTX 3060 12GBQ4_K_M · 8 GB · ~40 tok/s · 32K8~4032K12 GB tier →

16 GB · RTX 5060 Ti 16GBQ4_K_M · 8 GB · ~50 tok/s · 64K8~5064K16 GB tier →

24 GB · RTX 4090referenceQ4_K_M · 8 GB · ~112 tok/s · 64K8~11264K24 GB tier →

Mac 32 GB · Mac Mini M4Q4_K_M · 8 GB · ~14 tok/s · 128K8~14128KMacs →

Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 8 GB · ~65 tok/s · 256K8~65256KMacs →

Alternatives to AesCode 8B

Similar models

6tok/s on RTX 4090

Model	Score	VRAM	tok/s	Why it is similar
Granite Switch 4.1 8BIBM	—	20.7 GB	~36	same size
Apertus V1.5 8Bswiss-ai	—	8.5 GB	~111	same size
Ornith 1.0 9BDeep Reinforce	—	6.5 GB	~106	same size
LFM2.5 8B A1BLiquid AI	—	5.6 GB	~335	same size
MiniCPM O 4 5OpenBMB	—	8.3 GB	~105	same size
Qwen3 VL 8B InstructAlibaba	—	8 GB	~112	same size

Specifications

Publisher: Microsoft
Parameters: 8.8B
Architecture: Qwen3VLForConditionalGeneration
Context Length: 262,144 tokens
Vocabulary Size: 151,936
Release Date: 2026-09-29
License: Apache 2.0

Resources

Hugging Face model card (opens in a new tab)

Frequently Asked Questions

How much VRAM does AesCode 8B need?

AesCode 8B requires 5.9 GB of VRAM at Q4_K_M, or 18.1 GB at BF16. Full 262K context adds up to 38.4 GB (44.2 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 8.8B × 4.8 bits ÷ 8 = 5.3 GB

KV Cache + Overhead ≈ 0.6 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 38.9 GB (at full 262K context)

VRAM usage by quantization

Q4_K_M

5.9 GB

Q4_K_M + full context

44.2 GB

Learn more about VRAM estimation →

What's the best quantization for AesCode 8B?

For AesCode 8B, Q4_K_M (5.9 GB) offers the best balance of quality and VRAM usage. Q4_K_L (6.0 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 3.0 GB.

VRAM requirement by quantization

IQ2_XXS

3.0 GB~53%

IQ3_XS

4.2 GB~73%

Q4_0

5.0 GB~85%

Q4_K_M ★

5.9 GB~89%

Q4_K_L

6.0 GB~90%

BF16

18.1 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run AesCode 8B on a Mac?

Yes, but only at lower quantizations. The smallest Mac that can run AesCode 8B is Mac Mini M1 (8 GB) at IQ2_XXS; 156 of the 156 Macs we list can run it at some quantization. For Q4_K_M (5.9 GB) you need a Mac with more unified memory.

Can I run AesCode 8B locally?

Yes — AesCode 8B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is AesCode 8B?

At Q4_K_M, AesCode 8B can reach ~819 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~112 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = (bandwidth GB/s ÷ model GB) × efficiency

Example: NVIDIA B200 → 8000 ÷ 5.86 × 0.65 = ~887 tok/s

Estimated speed at Q4_K_M (5.9 GB)

NVIDIA B200

~887 tok/s

NVIDIA GeForce RTX 4090

~112 tok/s

NVIDIA B300

~887 tok/s

AMD Instinct MI350X

~819 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 AesCode 8B?

At Q4_K_M, the download is about 5.26 GB. The full-precision BF16 version is 17.53 GB. The smallest option (IQ2_XXS) is 2.41 GB.

Which GPUs can run AesCode 8B?

52 consumer GPUs can run AesCode 8B at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT, AMD Radeon RX 7600. 40 GPUs have plenty of headroom for comfortable inference.

Which devices can run AesCode 8B?

159 devices with unified memory can run AesCode 8B at Q4_K_M (5.9 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Apple iPhone 17 Pro, Asus ROG Flow Z13 (2025, Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.