the AI bench
VERIFIED JUNE 2026
All models

MODEL · NVLABS + MIT HAN LAB · 0.6B / 1.6B (LINEAR-ATTENTION DIT)

SANA (0.6B / 1.6B)

NVLabs + MIT Han Lab's linear-attention diffusion transformer. Fastest image generation at any given quality tier in its class — 23–39× faster than FLUX-dev on the same hardware.

License: NVIDIA NSCL v2 — non-commercial (easy misread as MIT; it is NOT) · Context: Up to 4K × 4K; sub-second 1024² on 16 GB GPU · Released: October 2024 (1.6B) / 2025 (Sprint variants)

The decision in five lines

The call
Consider — for image
Best for
image
Runs on
23 hardware picks fit (cheapest: Intel Arc B580 12 GB · $249)
Watch out
Also skip if absolute photorealism at complex compositions matters — linear attention trades some fidelity for speed.
Evidence
Estimated · last verified April 2026

0.6B
PARAMETERS
IMAGE GEN
TYPE
Up
CONTEXT
~2 GB (0.6B) / ~5 GB (1.6B)
VRAM AT Q4

Where we recommend this

Every tier slot in the planner where this model is a top or alternate pick. Pulled live from planner.js — when the planner refreshes, this table stays current.

IMAGE · MID
SANA-1.6B (non-commercial)NVIDIA 4K-capable 1.6B; extremely fast on mid-tier GPUs; weights are NVIDIA NSCL v2 (non-commercial).
IMAGE · LOW
SANA-0.6B (non-commercial)0.6B params; <1s per 1024² on a 16GB laptop GPU; weights are NVIDIA NSCL v2 (non-commercial).

The call

NVLabs + MIT Han Lab's linear-attention diffusion transformer. Fastest image generation at any given quality tier in its class — 23–39× faster than FLUX-dev on the same hardware.

When not to use: Anything commercial — NSCL v2 blocks it, period. Also skip if absolute photorealism at complex compositions matters — linear attention trades some fidelity for speed.

Runner notes

Diffusers `SanaPipeline` or ComfyUI with SANA custom nodes. Always disclose the NSCL restriction in anything you publish using it — users assume NVIDIA = permissive, which is wrong here.

License
NVIDIA NSCL v2 — non-commercial (easy misread as MIT; it is NOT)
Released
October 2024 (1.6B) / 2025 (Sprint variants)
Maker
NVLabs + MIT Han Lab

Hardware that fits

Every hardware pick whose memory fits this model at the quant we recommend. Sorted cheapest-first — the top row is your best-value fit. Click through for the full buyer’s guide.

Next step

Find-by-model — see what hardware runs this