dexiio

Best local LLM for vision / image input on a CPU only / integrated graphics (2026)

All figures assume an f16 KV cache, a 0.6 GB display reserve on the GPU, and 64 GB of DDR5 system RAM for the offload tiers. Tune these in the calculator.

The verdict

Gemma 3 27B Q8_0

CPU only at 16K · Vision / image input score 8/10 · ≈ 2 tok/s · needs 33 GB system RAM

Sliding-window attention keeps long context cheap, and the vision stack still beats most of 2026's newcomers.

llama-server -m gemma-3-27b-it-Q8_0.gguf -c 16384 --flash-attn -ngl 0

Worthy alternates

Gemma 4 26B A4B Q8_0

CPU only · ≈ 12 tok/s · Vision / image input 8/10

Google's fast MoE with native audio in. Nearly all of its weight sits in routed experts, so expert offload runs it comfortably on 12 GB cards.

Gemma 4 12B Q8_0

CPU only · ≈ 4 tok/s · Vision / image input 7/10

June 2026's 16GB-class headline: dense 12B with native vision and audio in one backbone. The new laptop ceiling.

Tune this for your exact RAM and settings in the calculator → · All models on the CPU only / integrated graphics

Frequently asked questions

What is the best local LLM for vision / image input on a CPU only / integrated graphics?

Gemma 3 27B at Q8_0 — it scores 8/10 for vision / image input and runs as "CPU only" at 16K context on the CPU only / integrated graphics.

How much context do I need for vision / image input?

We recommend 16K tokens for vision / image input (minimum 8K). These picks are computed at 16K.

How fast will it run on a CPU only / integrated graphics?

Roughly 2 tokens/sec for Gemma 3 27B — usable for interactive use.

Do I need more than 0 GB of VRAM for vision / image input?

No — the pick above needs 0 GB of VRAM plus 33 GB of system RAM at 16K.

What settings should I use?

Start with our command: llama-server -m gemma-3-27b-it-Q8_0.gguf -c 16384 --flash-attn -ngl 0 — then tune context and KV quant in the fit calculator.