Local AI Picksdata checked 2026-10-11

Local LLMs that fit in 16 GB

13 of 47 current models have a listed weight option that needs at most 85% of 16 GB.

against memory

OpenRouter evals, GPQA Diamond · data from 2026-10-10

0 of 15 models within 16 GB

No model with a score fits this memory budget.

Scores were measured on hosted endpoints, so each model is plotted once, at the memory need of its original weights. A quantized file needs less memory and may score differently. Lines connect the frontier points and do not predict results in between. Source: OpenRouter evals (openrouter.ai) via OpenRouter (openrouter.ai/rankings). openrouter.ai

Models

13
ModelLabParamsLargest option that fitsScore
FrogNano 4BMicrosoft4.66B4B Original weights · 10.2 GBtiny-aya ThinkerPreviewCohere3.35BOriginal weights · 6.8 GBGranite 4.2IBM3.66B30B GGUF Q2_K · 12.7 GBLing 3.0Ant Group inclusionAI7.89BGGUF Q8_0 · 9.9 GBNorth Micro VisionPreviewCohere2.48BOriginal weights · 6.1 GBMage-VLMicrosoft4.74BOriginal weights · 11.5 GBGranite SwashIBM2.14B2B Original weights · 4.9 GBUltraX 0.6B PreviewOpenBMB—0.6B Original weights · 3.9 GBMiniCPM5OpenBMB1.08B2B GGUF F16 · 5.6 GBBitCPM-CANNOpenBMB—3B GGUF BF16 · 7.5 GBGranite Switch 4.1 PreviewIBM4.15B3B Original weights · 8.9 GBnanowhale 100MHugging Face110M100M Original weights · 1.0 GBMiniCPM-V 4.6OpenBMB1.3BOriginal weights · 3.3 GB

Models with a GPQA Diamond score come first, highest score first. The rest are newest first.

An option is listed when its memory need is at most 85% of 16 GB. The figure after each option is an estimate from file size, an 8K-token context and 0.5 GB for the runtime, unless the publisher states one.