TokenAssemble

Google · open-weights model

gemma-4-26B-A4B-it

gemma-4-26B-A4B-it is a 26.5B-parameter model (MoE, 3.8B active) by Google. Its smallest quant (MXFP4_MOE) needs at least 19.0GB of GPU-addressable memory at 8K context; it loads cleanly at a quality quant on 20 of the 50 machines we track.

Parameters
26.5B total · 3.8B active
Layers
30
Max context
256K
Attention
HYBRID
License
apache-2.0

Config from google/gemma-4-26B-A4B-it

Quants and what they need

Measured file sizes; “needs” = weights + KV cache at 8K context + overhead.

QuantFile sizeNeeds ≥
Q8_026.9 GB29.3 GB
UD-Q5_K_M21.1 GB23.6 GB
UD-Q4_K_M16.9 GB19.4 GB
MXFP4_MOE16.6 GB19.0 GB

What runs gemma-4-26B-A4B-it

Every machine we track, graded at its best clean-fit quant.

HardwareBest-quant verdict · llama.cpp · 8K context
Intel Arc A770 16 GBTight · C · MXFP4_MOEinteractive · offloadbeta
Intel Arc B580Tight · C · MXFP4_MOEinteractive · offloadbeta
AMD Radeon PRO W7900Fits · A+ · Q8_0interactivebeta
NVIDIA GeForce RTX 3060 12 GBTight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 3070Tight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 3070 TiTight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 3080 10 GBTight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 3080 12 GBTight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 3080 TiTight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 3090Fits · B · UD-Q4_K_Minteractivebeta
NVIDIA GeForce RTX 3090 TiFits · B · UD-Q4_K_Minteractivebeta
NVIDIA GeForce RTX 4060Tight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 4060 Ti 16 GBTight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 4060 Ti 8 GBTight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 4070Tight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 4070 SUPERTight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 4070 TiTight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 4070 Ti SUPERTight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 4080Tight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 4080 SUPERTight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 4090Fits · B · UD-Q4_K_Minteractivebeta
NVIDIA RTX 5000 Ada GenerationFits · B · UD-Q5_K_Minteractivebeta
NVIDIA GeForce RTX 5060 Ti 16 GBTight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 5070Tight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 5070 TiTight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 5080Tight · C · MXFP4_MOEinteractive · offloadbeta
NVIDIA GeForce RTX 5090Fits · B · UD-Q5_K_Minteractivebeta
NVIDIA RTX 6000 Ada GenerationFits · A+ · Q8_0interactivebeta
NVIDIA RTX A6000Fits · A+ · Q8_0interactivebeta
AMD Radeon RX 6800Tight · C · MXFP4_MOEinteractive · offloadbeta
AMD Radeon RX 6900 XTTight · C · MXFP4_MOEinteractive · offloadbeta
AMD Radeon RX 7800 XTTight · C · MXFP4_MOEinteractive · offloadbeta
AMD Radeon RX 7900 XTTight · C · MXFP4_MOEinteractivebeta
AMD Radeon RX 7900 XTXFits · B · UD-Q4_K_Minteractivebeta
AMD Radeon RX 9070Tight · C · MXFP4_MOEinteractive · offloadbeta
AMD Radeon RX 9070 XTTight · C · MXFP4_MOEinteractive · offloadbeta
ASUS Ascent GX10Fits · A+ · Q8_0interactivebeta
Beelink GTR9 Pro Ryzen AI Max+ 395 128GBFits · A+ · Q8_0usablebeta
NVIDIA DGX SparkFits · A+ · Q8_0interactivebeta
Framework Desktop Ryzen AI Max+ 395 128GBFits · A+ · Q8_0usablebeta
Framework Desktop Ryzen AI Max+ 395 64GBFits · A+ · Q8_0usablebeta
GMKtec EVO-X2 Ryzen AI Max+ 395 128GBFits · A+ · Q8_0usablebeta
HP Z2 Mini G1a Ryzen AI Max+ PRO 395 128GBFits · A+ · Q8_0usablebeta
Mac mini M4 16GBWon't fitbeta
Mac mini M4 24GBWon't fitbeta
Mac mini M4 Pro 24GBWon't fitbeta
Mac mini M4 Pro 48GBFits · B · Q8_0usablebeta
Mac Studio M3 Ultra 96GBFits · A+ · Q8_0interactivebeta
Mac Studio M4 Max 36GBFits · B · UD-Q5_K_Minteractivebeta
Mac Studio M4 Max 64GBFits · A+ · Q8_0interactivebeta

Verdicts computed from published specs and measured GGUF sizes — how the math works.