Advisor / Models / Gemma 4 26B-A4B

Gemma 4 26B-A4B: VRAM requirements & speed

Gemma 4 26B-A4B (25.81B MoE, 3.8B active, apache-2.0). Native context 256k. Smallest GGUF: 16.95 GB.

16.95 GBsmallest quant file
18hardware pages that fit it fully
7run with offload
3measured data points

Memory needed by quantization and context

GGUF file size plus f16 KV cache (and fixed overhead of ~1 GB per GPU not shown). Source: unsloth/gemma-4-26B-A4B-it-GGUF.

QuantFile+8k ctx+32k ctx+128k ctx+256k ctx
Q4_K_M UD-Q4_K_M16.95 GB17.3 GB17.8 GB19.8 GB22.5 GB
Q5_K_M UD-Q5_K_M21.15 GB21.5 GB22.0 GB24.0 GB26.7 GB
Q6_K UD-Q6_K23.17 GB23.5 GB24.1 GB26.1 GB28.7 GB
Q8_027.32 GB27.7 GB28.2 GB30.2 GB32.9 GB

KV cache size by context length

Computed from the model's attention layout (sliding-window layers capped at their window).

Contextf16q8_0q4_0
8k0.38 GB0.20 GB0.11 GB
32k0.88 GB0.47 GB0.25 GB
128k2.89 GB1.54 GB0.81 GB
256k5.58 GB2.96 GB1.57 GB

Which hardware runs Gemma 4 26B-A4B? · 16k context

HardwareQuantFitMemoryTok/sSpeed source
RTX PRO 6000 96GB (bonus page)Q8_0fits fully27.9 GB215estimated
RTX 5090 32GBQ8_0fits fully27.9 GB213scaled from measured
Mac Studio M5 Ultra 256GBQ8_0fits fully27.9 GB148estimated
Mac Studio M5 Ultra 512GB (ships late Oct 2026)Q8_0fits fully27.9 GB148estimated
AMD RX 7900 XTX 24GBUD-Q5_K_Mfits fully21.7 GB147estimated
RTX 4090 24GBUD-Q5_K_Mfits fully21.7 GB135scaled from measured
2x RTX 3090 (48GB total)Q8_0fits fully27.9 GB125estimated
RTX 3090 24GBUD-Q5_K_Mfits fully21.7 GB112scaled from measured
RTX 5080 16GBUD-Q4_K_MMoE offload to CPU17.5 GB102estimated
RTX 5070 Ti 16GBUD-Q4_K_MMoE offload to CPU17.5 GB98estimated
16GB VRAM class (RTX 4060 Ti 16GB, 32GB RAM)UD-Q4_K_MMoE offload to CPU17.5 GB68estimated
Mac Studio (2025) M3 Ultra 96GBQ8_0fits fully27.9 GB63estimated
Mac Studio M5 Max 128GBQ8_0fits fully27.9 GB57scaled from measured
MacBook Pro 14/16 M5 Max 128GBQ8_0fits fully27.9 GB57scaled from measured
NVIDIA DGX Spark 128GBQ8_0fits fully27.9 GB53scaled from measured
MacBook Pro 14/16 (2024) M4 Max 128GBQ8_0fits fully27.9 GB50estimated
RTX 5070 12GBUD-Q4_K_MMoE offload to CPU17.5 GB48estimated
RTX 4070 12GBUD-Q4_K_MMoE offload to CPU17.5 GB46estimated
MacBook Pro 14/16 M5 Pro 48GBQ8_0fits fully27.9 GB42estimated
AMD Ryzen AI Max+ 395 (Strix Halo) 128GBQ8_0fits fully27.9 GB35scaled from measured
Mac mini (2024) M4 Pro 64GBQ8_0fits fully27.9 GB31estimated
8GB VRAM class (RTX 4060 8GB, 32GB RAM)UD-Q4_K_MMoE offload to CPU17.5 GB29estimated
RTX 3060 12GB (32GB RAM)UD-Q4_K_MMoE offload to CPU17.5 GB27estimated
Mac mini M6 32GBUD-Q5_K_Mfits fully21.7 GB27estimated
MacBook Air M5 32GBUD-Q5_K_Mfits fully21.7 GB25estimated
Mac mini / iMac M4 16GB-does not fit--

Discrete GPUs are modelled with 32 GB of system RAM available for offload. Does not fit even with offload: Mac mini / iMac M4 16GB.

Links

Model card · GGUF files · Open in the advisor · last verified 2026-10-14

FAQ

How much VRAM does Gemma 4 26B-A4B need?

The smallest quant (Q4_K_M) is 16.95 GB, so you need roughly 18 GB for an 8k context including KV cache and overhead. At 256k context the f16 KV cache alone is 5.6 GB.

Can I run Gemma 4 26B-A4B on a RTX PRO 6000 96GB (bonus page)?

Yes: on RTX PRO 6000 96GB (bonus page) it fits fully as Q8_0 at about 215 tok/s (estimated). 18 of 26 hardware configurations fit it fully at 16k context.

What is the maximum context for Gemma 4 26B-A4B?

256k tokens natively. Memory, not the model, is usually the limit: see the KV cache table above.

All models: model index · by memory: 8 GB · 12 GB · 16 GB · 24 GB · 32 GB · 48 GB · 64 GB · 96 GB · 128 GB · 192 GB · 256 GB · 512 GB. Numbers use the same fit math as the advisor; sources and methods.