All tracked LLMs

97 open-weight models with benchmarks and hardware requirements.

This is the complete catalogue of open-weight large language models tracked by CanItRun. For each model you can see the exact VRAM needed at every quantization level (FP16 down to Q2_K), the estimated inference speed in tokens per second on any GPU we track, and benchmark scores from the Open LLM Leaderboard v2 and LMSYS Chatbot Arena.

Models range from sub-1B edge models to 1T+ parameter frontier models. Use the homepage calculator to filter by your specific hardware, or browse below to compare models by parameter count. Each model page shows the full compatibility breakdown across all {60+} GPUs we track.

Browse by size

License

Releaser

Sort

97 models

Guides by use case

Each of these goes deeper than the filters above: VRAM breakdowns, GPU pairings, and tradeoffs specific to one category of model.

Coding LLMs

31 models

Best open-weight coding models for local AI: VRAM requirements, benchmarks, and compatible GPUs for running coding assistants offline.

Reasoning LLMs

44 models

Best open-weight reasoning models for local AI: chain-of-thought, math, and complex problem-solving with VRAM requirements and GPU compatibility.

Multimodal LLMs

33 models

Best open-weight multimodal models for local AI: vision, audio, and video support with VRAM requirements and compatible GPUs.

Small LLMs

11 models

Best small open-weight LLMs for local AI: models under 10B parameters that run on 8 GB or 16 GB GPUs with VRAM requirements and benchmarks.

Mixture of Experts LLMs

46 models

Best open-weight Mixture of Experts (MoE) LLMs for local AI: high capability with fewer active parameters, VRAM requirements, and GPU compatibility.

General-Purpose LLMs

30 models

Best open-weight general-purpose LLMs for local AI: strong chat, instruction-following, and reasoning with VRAM requirements and GPU compatibility.

Frontier LLMs

40 models

Open-weight frontier models for local AI: the largest and most capable models, VRAM requirements, and multi-GPU compatibility.