utGoogle · Downloadable
t5gemma-xl-xl-prefixlm-it
Generates text for conversations and writing.
- Source listed
- Jun 19, 2025
- Parameters
- 3.8B
- 4-bit weights only
- ≈ 1.9 GB
- License
- gemma
YOUR MEMORY BUDGET
Local AI models for 8GB VRAM.
Explore models by theoretical weight storage. This shortlist covers available downloadable language models with a known or name-derived parameter count.
6 GBleft for weights at 4-bit
8 GB budget − 2 GB reserve
8 GB budget − 2 GB reserve
The reserve is your assumption, not a measured overhead allowance. The default 2 GB may be too little for your runtime or context length. Results are not confirmed fits and have no verified speed rating. Weight estimates use decimal GB and do not include quantization overhead. For timed results on one 12GB machine, see measured speeds.
852 models within the weight budget
Largest parameter count first · not a quality rankingGoogle · Downloadable
t5gemma-xl-xl-ul2
Generates text for conversations and writing.
- Source listed
- Jun 19, 2025
- Parameters
- 3.8B
- 4-bit weights only
- ≈ 1.9 GB
- License
- gemma
Google · Downloadable
t5gemma-xl-xl-ul2-it
Generates text for conversations and writing.
- Source listed
- Jun 19, 2025
- Parameters
- 3.8B
- 4-bit weights only
- ≈ 1.9 GB
- License
- gemma
Microsoft · Downloadable
Phi-tiny-MoE-instruct
Generates text for conversations and writing.
- Source listed
- Jun 23, 2025
- Parameters
- 3.8B
- 4-bit weights only
- ≈ 1.9 GB
- License
- mit
StepFun · Downloadable
Step-Audio-EditX
Generates text, according to its published model architecture (Step1ForCausalLM). No task is listed on Hugging Face, so check the model card for intended use.
- Source listed
- Oct 29, 2025
- Parameters
- 3.5B
- 4-bit weights only
- ≈ 1.8 GB
- License
- Not disclosed
NVIDIA · Downloadable
Kimi-K2.6-DFlash
Generates text for conversations and writing.
- Source listed
- Jun 23, 2026
- Parameters
- 3.5B
- 4-bit weights only
- ≈ 1.7 GB
- License
- other
NVIDIA · Downloadable
Kimi-K2.7-Code-DFlash
Designed for coding and software tasks. Generates text for conversations and writing.
- Source listed
- Jul 8, 2026
- Parameters
- 3.5B
- 4-bit weights only
- ≈ 1.7 GB
- License
- other
DeepSeek · Downloadable
DeepSeek-OCR-2
Understands text and images and responds in text.
- Source listed
- Jan 27, 2026
- Parameters
- 3.4B
- 4-bit weights only
- ≈ 1.7 GB
- License
- apache-2.0
DeepSeek · Downloadable
deepseek-vl2-tiny
Understands text and images and responds in text.
- Source listed
- Dec 13, 2024
- Parameters
- 3.4B
- 4-bit weights only
- ≈ 1.7 GB
- License
- other
Baidu · Downloadable
Unlimited-OCR
Understands text and images and responds in text.
- Source listed
- Jun 19, 2026
- Parameters
- 3.3B
- 4-bit weights only
- ≈ 1.7 GB
- License
- mit
DeepSeek · Downloadable
DeepSeek-OCR
Understands text and images and responds in text.
- Source listed
- Oct 17, 2025
- Parameters
- 3.3B
- 4-bit weights only
- ≈ 1.7 GB
- License
- mit
NVIDIA · Downloadable
MiniMax-M3-DSpark
Generates text for conversations and writing.
- Source listed
- Jul 15, 2026
- Parameters
- 3.3B
- 4-bit weights only
- ≈ 1.6 GB
- License
- other
Aleph Alpha · Downloadable
sp-baseline-research-3b
Generates text for conversations and writing.
- Source listed
- Oct 29, 2024
- Parameters
- 3B
- 4-bit weights only
- ≈ 1.5 GB
- License
- other
Aleph Alpha · Downloadable
umup-research-3b
Generates text for conversations and writing.
- Source listed
- Oct 29, 2024
- Parameters
- 3B
- 4-bit weights only
- ≈ 1.5 GB
- License
- other
BAAI · Downloadable
Bunny-v1_0-3B
Generates text for conversations and writing.
- Source listed
- Feb 20, 2024
- Parameters
- 3B
- 4-bit weights only
- ≈ 1.5 GB
- License
- apache-2.0
BAAI · Downloadable
Bunny-v1_0-3B-zh
Generates text for conversations and writing.
- Source listed
- Apr 18, 2024
- Parameters
- 3B
- 4-bit weights only
- ≈ 1.5 GB
- License
- apache-2.0
Baidu · Downloadable
Qianfan-VL-3B
Understands text and images and responds in text.
- Source listed
- Sep 16, 2025
- Parameters
- 3B
- 4-bit weights only
- ≈ 1.5 GB
- License
- other
Kyutai · Downloadable
CASA-Qwen2_5-VL-3B
Understands text and images and responds in text.
- Source listed
- Dec 10, 2025
- Parameters
- 3B
- 4-bit weights only
- ≈ 1.5 GB
- License
- cc-by-nc-sa-4.0
Kyutai · Downloadable
CASA-Qwen2_5-VL-3B-LiveCC
Generates text for conversations and writing.
- Source listed
- Dec 18, 2025
- Parameters
- 3B
- 4-bit weights only
- ≈ 1.5 GB
- License
- cc-by-nc-sa-4.0
Kyutai · Downloadable
CASA-Qwen2_5-VL-3B-Shared
Understands text and images and responds in text.
- Source listed
- Aug 26, 2026
- Parameters
- 3B
- 4-bit weights only
- ≈ 1.5 GB
- License
- cc-by-nc-sa-4.0
Liquid AI · Downloadable
d1-3B
Understands text and images and responds in text.
- Source listed
- Oct 5, 2026
- Parameters
- 3B
- 4-bit weights only
- ≈ 1.5 GB
- License
- other
Liquid AI · Downloadable
LFM2-VL-3B
Understands text and images and responds in text.
- Source listed
- Oct 22, 2025
- Parameters
- 3B
- 4-bit weights only
- ≈ 1.5 GB
- License
- other
Liquid AI · Downloadable
LFM2.5-VL-3B
Understands text and images and responds in text.
- Source listed
- Aug 11, 2026
- Parameters
- 3B
- 4-bit weights only
- ≈ 1.5 GB
- License
- other
Liquid AI · Downloadable
LFM2.5-VL-3B-DSpark
Understands text and images and responds in text.
- Source listed
- Sep 18, 2026
- Parameters
- 3B
- 4-bit weights only
- ≈ 1.5 GB
- License
- other
What changes with a 8GB budget?
With 2 GB set aside, 6 GB remains for weights. At 4-bit that allows up to 12 billion parameters in the theoretical calculation. An 8-billion-parameter model alone needs approximately 4 GB at this precision, before its conversation cache and runtime.
Why might a listed model still fail to load?
Its actual weight format may need more space, the runtime may not support its architecture, or the conversation cache may exceed your reserve. Vision models can also need space for image processing. Check the original model card and measure your exact setup before relying on it.
Read the memory guide and technical sources · Compare two models