Best local LLM for 8GB VRAM: what fits (moved to our fuller guide)

There's a newer, fuller guide on this: Only 8GB of VRAM? The local AI models that actually fit.

The short version, as of November 1, 2026: on our VRAM numbers, Qwen3.5 9B fits an 8GB card at Q5_K_M with 8K tokens of context (about 7.4 GB) and at Q4_K_M with 32K (about 7.2 GB). Gemma 4 E4B at Q6_K (about 7.0 GB at 8K) is the pick if you want headroom. Older advice that put Qwen3 8B or Llama 3.1 8B at the top is out of date: newer hybrid-attention models hold far more context in the same memory.

Bigger models such as Qwen3.8 Flash Next can run on an 8GB card only when system RAM holds most of the model, at a 2-bit or 3-bit quant and slower. The updated guide covers that in full, and the VRAM ladder compares every tier.