Best local AI models for NVIDIA GTX 1650

The largest models that fit fully

The 30 largest of the 81 models that fit; every smaller model in the catalog fits too. Best quant means the highest quality compression whose weights and 4k context both sit inside the memory.

Close, but only with CPU offload

These need more than the card holds at their smallest practical quant, so part of the model runs from system memory (figures assume 32 GB of it). They work, several times slower.

How to read this

The NVIDIA GTX 1650 graphics card features 4 GB GDDR5 memory. This memory size determines which local artificial intelligence models can run directly on your hardware. To fit within this limit, models must use quantization. Quantization is a compression method that reduces the size of a model while preserving its capabilities. The best quant column indicates the highest quality compression level that safely fits inside the available video memory.

For maximum performance, you can run several models entirely on the graphics card. The largest fitting models include Lumina-Next or Lumina-Image 2.0 and CogVideoX 2B or 5B. Both are 5B models using the Q4_K_M quant and consuming 3.7 GB of video memory. DeepSeek-VL2 is a 4.5B model that fits at the Q5_K_M quant using 3.8 GB. DeepFloyd IF is a 4.3B model using 3.7 GB at Q5_K_M. Phi-3.5-vision is a 4.2B model using 3.6 GB at Q5_K_M.

Several 4B models fit using the Q6_K quant which requires 3.9 GB of memory. These models include Qwen3 4B, Gemma 3 4B, Gemma 4 E4B, MiniCPM 3 4B, Danube 3 4B, and Fish Speech 1.5 or OpenAudio S1. The 3.8B models like Phi-4-mini-instruct, Phi-3.5 Mini, and OmniGen or OmniGen2 also run at Q6_K using 3.7 GB of memory. Image generation models like SD Cascade (Würstchen v3) at 3.6B use 3.5 GB, while SDXL Turbo and SDXL Lightning at 3.5B use 3.4 GB at Q6_K. ACE-Step at 3.5B also uses 3.4 GB, and MusicGen small/medium/large at 3.3B uses 3.2 GB.

Smaller models can run at the higher quality Q8_0 quant. The 3B models like SmolLM3 3B, Replit Code v1.5 3B, Kandinsky 3.1, Voxtral Mini or Small, Orpheus TTS, and Higgs Audio v2 use 3.8 GB of video memory. Video and image models also fit at Q8_0. Allegro at 2.8B uses 3.6 GB. Open-Sora Plan at 2.7B uses 3.4 GB. LFM2 1.2B or 2.6B and Playground v2.5 at 2.6B use 3.3 GB. Stable Diffusion 3.5 Medium at 2.5B uses 3.2 GB.

If a model is too large for the 4 GB video memory, you can offload parts of it to your system RAM. This offload process requires a system with 32 GB RAM. Offloading allows you to run larger models but reduces processing speed because system RAM is slower than video memory. For example, Stable Diffusion XL is a 3.417B model that needs 4.1 GB at FP8 or optimized settings and uses 6.1 GB of system RAM. Phi-3 Mini is a 3.8B model that needs 4.4 GB at Q4_K_M and uses 6.4 GB of system RAM. Phi-4-multimodal is a 5.6B model needing 4.1 GB at Q4_K_M and 6.1 GB of system RAM.

You can also run larger 7B class models using CPU offload. Magicoder-S-DS 6.7B needs 4.9 GB at Q4_K_M and 6.9 GB of system RAM. Mistral 7B needs 5.7 GB at Q4_K_M and 7.7 GB of system RAM. Several other models need 5.1 GB at Q4_K_M and 7.1 GB of system RAM. These models include Qwen2.5 0.5B or 1.5B or 3B or 7B, OLMo 2 1B or 7B, Falcon 3 1B or 3B or 7B, Command R7B, and OpenHermes 2.5.

When running these models, you must consider the context limit. The memory calculations shown here assume a standard 4k context window. If you increase the context length to process longer documents, the model will require more memory. This extra memory usage might exceed the 4 GB limit of your graphics card and force the system to slow down.