utDatalab · Downloadable
surya-parsebench-rl
Understands text and images and responds in text.
- Source listed
- Oct 12, 2026
- Parameters
- 665.7M
- 4-bit weights only
- ≈ 0.3 GB
- License
- openrail
YOUR MEMORY BUDGET
Local AI models for 12GB VRAM.
Explore models by theoretical weight storage. This shortlist covers available downloadable language models with a known or name-derived parameter count.
10 GBleft for weights at 4-bit
12 GB budget − 2 GB reserve
12 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.
980 models within the weight budget
Largest parameter count first · not a quality rankingMeta · Downloadable
MobileMoE-S-QAT
Generates text for conversations and writing.
- Source listed
- Jun 11, 2026
- Parameters
- 665.3M
- 4-bit weights only
- ≈ 0.3 GB
- License
- fair-noncommercial-research-license
NVIDIA · Downloadable
Nemotron-3.5-Parse
Understands text and images and responds in text.
- Source listed
- Sep 3, 2026
- Parameters
- 630.8M
- 4-bit weights only
- ≈ 0.3 GB
- License
- openmdw-1.1
Liquid AI · Downloadable
d1-omni-600M
Understands text and images and responds in text.
- Source listed
- Oct 5, 2026
- Parameters
- 600M
- 4-bit weights only
- ≈ 0.3 GB
- License
- other
NVIDIA · Downloadable
Privasis-Cleaner-0.6B
Generates text for conversations and writing.
- Source listed
- Jun 8, 2026
- Parameters
- 600M
- 4-bit weights only
- ≈ 0.3 GB
- License
- other
Qwen · Alibaba · Downloadable
Qwen3-0.6B
Generates text for conversations and writing.
- Source listed
- Apr 27, 2025
- Parameters
- 600M
- 4-bit weights only
- ≈ 0.3 GB
- License
- apache-2.0
Qwen · Alibaba · Downloadable
Qwen3Guard-Gen-0.6B
Generates text for conversations and writing.
- Source listed
- Sep 23, 2025
- Parameters
- 600M
- 4-bit weights only
- ≈ 0.3 GB
- License
- apache-2.0
Google · Downloadable
t5gemma-b-b-prefixlm
Generates text for conversations and writing.
- Source listed
- Jun 19, 2025
- Parameters
- 591.5M
- 4-bit weights only
- ≈ 0.3 GB
- License
- gemma
Google · Downloadable
t5gemma-b-b-prefixlm-it
Generates text for conversations and writing.
- Source listed
- Jun 19, 2025
- Parameters
- 591.5M
- 4-bit weights only
- ≈ 0.3 GB
- License
- gemma
Google · Downloadable
t5gemma-b-b-ul2
Generates text for conversations and writing.
- Source listed
- Jun 19, 2025
- Parameters
- 591.5M
- 4-bit weights only
- ≈ 0.3 GB
- License
- gemma
Google · Downloadable
t5gemma-b-b-ul2-it
Generates text for conversations and writing.
- Source listed
- Jun 19, 2025
- Parameters
- 591.5M
- 4-bit weights only
- ≈ 0.3 GB
- License
- gemma
StepFun · Downloadable
GOT-OCR-2.0-hf
Understands text and images and responds in text.
- Source listed
- Nov 22, 2024
- Parameters
- 560.5M
- 4-bit weights only
- ≈ 0.3 GB
- License
- apache-2.0
M-A-P · Downloadable
MuPT-v1-8192-550M
Generates text for conversations and writing.
- Source listed
- Jan 18, 2024
- Parameters
- 550M
- 4-bit weights only
- ≈ 0.3 GB
- License
- apache-2.0
NVIDIA · Downloadable
NVIDIA-Nemotron-3.5-Bolt-E4B-BF16-DFlash
Generates text for conversations and writing.
- Source listed
- Oct 9, 2026
- Parameters
- 515.8M
- 4-bit weights only
- ≈ 0.3 GB
- License
- other
M-A-P · Downloadable
YuE-s1-0.5B
Generates text for conversations and writing.
- Source listed
- Mar 18, 2025
- Parameters
- 500M
- 4-bit weights only
- ≈ 0.3 GB
- License
- apache-2.0
OpenBMB · Downloadable
BitCPM-CANN-0.5B
Generates text for conversations and writing.
- Source listed
- May 15, 2026
- Parameters
- 500M
- 4-bit weights only
- ≈ 0.3 GB
- License
- apache-2.0
OpenBMB · Downloadable
BitCPM4-0.5B
Generates text for conversations and writing.
- Source listed
- Jun 5, 2025
- Parameters
- 500M
- 4-bit weights only
- ≈ 0.3 GB
- License
- apache-2.0
OpenBMB · Downloadable
MiniCPM4-0.5B
Generates text for conversations and writing.
- Source listed
- Jun 5, 2025
- Parameters
- 500M
- 4-bit weights only
- ≈ 0.3 GB
- License
- apache-2.0
OpenBMB · Downloadable
MiniCPM4-0.5B-QAT-Int4-GPTQ-format
Generates text for conversations and writing.
- Source listed
- Jun 9, 2025
- Parameters
- 500M
- 4-bit weights only
- ≈ 0.3 GB
- License
- apache-2.0
Tencent · Downloadable
Hunyuan-0.5B-Instruct
Generates text for conversations and writing.
- Source listed
- Jul 30, 2025
- Parameters
- 500M
- 4-bit weights only
- ≈ 0.3 GB
- License
- Not disclosed
Tencent · Downloadable
Hunyuan-0.5B-Pretrain
Generates text for conversations and writing.
- Source listed
- Jul 28, 2025
- Parameters
- 500M
- 4-bit weights only
- ≈ 0.3 GB
- License
- Not disclosed
inclusionAI · Downloadable
Ling-3.0-flash-UBTree
Generates text for conversations and writing.
- Source listed
- Oct 14, 2026
- Parameters
- 482M
- 4-bit weights only
- ≈ 0.2 GB
- License
- mit
Liquid AI · Downloadable
LFM2-VL-450M
Understands text and images and responds in text.
- Source listed
- Aug 12, 2025
- Parameters
- 450M
- 4-bit weights only
- ≈ 0.2 GB
- License
- other
Liquid AI · Downloadable
LFM2.5-VL-450M
Understands text and images and responds in text.
- Source listed
- Apr 8, 2026
- Parameters
- 450M
- 4-bit weights only
- ≈ 0.2 GB
- License
- other
What changes with a 12GB budget?
With 2 GB set aside, 10 GB remains for weights. At 4-bit that allows up to 20 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