How much VRAM does Gemma 4 26B A4B need?
Gemma 4 26B A4B needs 14.6 GB for its weights at 4-bit, 15.9 GB with an 8K context and 16.1 GB with 32K. At 8-bit, 26.8 GB.
- Parameters
- 25.8B
- Active per token
- 3.8B
- Longest context
- 256K
- Released
- Mar 2026
- License
- Apache 2.0
- Strongest at
- Chat, Coding, Agents, Docs
Memory at each precision
| Precision | Weights | With 8K | With 32K | With 128K |
|---|---|---|---|---|
| 16-bit | 48.1 GB | 49.4 GB | 49.6 GB | 50.6 GB |
| 8-bit | 25.5 GB | 26.8 GB | 27.1 GB | 28.0 GB |
| 6-bit | 19.7 GB | 21.0 GB | 21.3 GB | 22.2 GB |
| 5-bit | 17.1 GB | 18.4 GB | 18.6 GB | 19.6 GB |
| 4-bit | 14.6 GB | 15.9 GB | 16.1 GB | 17.1 GB |
| 3-bit | 11.7 GB | 13.1 GB | 13.3 GB | 14.2 GB |
| 2-bit | 9.0 GB | 10.3 GB | 10.6 GB | 11.5 GB |
| 1.6-bit | 5.0 GB | 6.3 GB | 6.5 GB | 7.4 GB |
One conversation, 16-bit context cache, runtime overhead included.
What runs it
| Machine | Best fit | Context | Speed | First reply |
|---|---|---|---|---|
| RTX 3060 12 GB | Too big | |||
| RTX 4060 Ti 16 GB | 3-bit | 224K | 91 tok/s | 1.1 s |
| RTX 3090 | 6-bit | 184K | 153 tok/s | 0.7 s |
| RTX 4090 | 6-bit | 184K | 161 tok/s | 0.3 s |
| RTX 5090 | 8-bit | 256K | 199 tok/s | 0.2 s |
| 2× RTX 3090 | 8-bit | 256K | 126 tok/s | 0.4 s |
| RTX PRO 6000 Blackwell | 8-bit | 256K | 199 tok/s | 0.2 s |
| M4 Max 128 GB | 8-bit | 256K | 64 tok/s | 3.7 s |
| M5 Max 128 GB | 8-bit | 256K | 70 tok/s | 1.1 s |
| Ryzen AI Max+ 395 128 GB | 8-bit | 256K | 34 tok/s | 2.1 s |
| DGX Spark 128 GB | 8-bit | 256K | 44 tok/s | 0.7 s |
| M3 Ultra 256 GB | 8-bit | 256K | 85 tok/s | 2.1 s |
| M5 Ultra 512 GB | 8-bit | 256K | 107 tok/s | 0.6 s |
| H100 SXM | 8-bit | 256K | 284 tok/s | 0.1 s |
| DGX Station GB300 | 8-bit | 256K | 383 tok/s | 0.1 s |
| 8× H200 NVL | 8-bit | 256K | 333 tok/s | 0.1 s |
Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.
How to run it
- Ollama
ollama run gemma4:26b- llama.cpp
llama-server -hf unsloth/gemma-4-26B-A4B-it-GGUF- vLLM
vllm serve google/gemma-4-26B-A4B-it- MLX (Mac)
mlx_lm.chat --model mlx-community/gemma-4-26b-a4b-it-4bit
Questions
How much VRAM does Gemma 4 26B A4B need?
Gemma 4 26B A4B needs 14.6 GB for its weights at 4-bit, 15.9 GB with an 8K context and 16.1 GB with 32K. At 8-bit, 26.8 GB.
Can Gemma 4 26B A4B run on an RTX 4090?
Yes. At 6-bit it holds 184K of context and writes about 161 tokens a second.