Google
Gemma 4 26B-A4B
A fast mixture-of-experts model. No preset yet: run it with vLLM on a base image, or call it by API.
Recommended: RTX 5090, 32 GB GDDR7, $0.69 / hour, billed per minute. Check availability.
Recommended: DGX Spark, 128 GB unified, $0.79 / hour, billed per minute. Check availability.
Recommended: DGX Spark, 128 GB unified, $0.79 / hour, billed per minute. Check availability.
Recommended: DGX Spark, 128 GB unified, $0.79 / hour, billed per minute. Check availability.
Configure
Every build below is published on Hugging Face and fits at least one GPU we rent. Pick the quantization and engine: the memory it needs, the GPUs it fits and the recommended GPU follow, and Deploy takes that exact build, revision and GPU to the console. Only the one-click preset is tested by GPUwerk.
| Quantization | NVFP4 |
|---|---|
| Engine | vLLM |
| Status | Untested by GPUwerk |
| Weights | 18.8 GB |
| GPU memory it needs | About 24.6 GB with a 32k-token cache, an estimate |
| Recommended GPU | RTX 5090, $0.69 / hour |
| Weights source | nvidia/Gemma-4-26B-A4B-NVFP4, revision a19cfe00be84 |
| Licence on the card | apache-2.0 |
| GPU | Memory | Price | Availability | Deploy |
|---|---|---|---|---|
| RTX 5090x86 · Blackwell · recommended | 32 GB GDDR7 | $0.69 / hour | Check availability | Run it yourself Gemma 4 26B-A4B NVFP4 on RTX 5090 |
| DGX SparkArm · GB10 | 128 GB unified | $0.79 / hour | Check availability | Run it yourself Gemma 4 26B-A4B NVFP4 on DGX Spark |
| RTX PRO 6000x86 · Blackwell | 96 GB GDDR7 | $1.89 / hour | Check availability | Run it yourself Gemma 4 26B-A4B NVFP4 on RTX PRO 6000 |
| H100 SXMx86 · Hopper | 80 GB HBM3 | Does not fit: NVFP4 weights need a Blackwell GPU; H100 is not one. | ||
| H200x86 · Hopper | 141 GB HBM3e | Does not fit: NVFP4 weights need a Blackwell GPU; H200 is not one. | ||
| B200x86 · Blackwell | 180 GB HBM3e | $5.49 / hour | Check availability | Run it yourself Gemma 4 26B-A4B NVFP4 on B200 |
| B300x86 · Blackwell | 288 GB HBM3e | $6.49 / hour | Check availability | Run it yourself Gemma 4 26B-A4B NVFP4 on B300 |
| Quantization | BF16 |
|---|---|
| Engine | vLLM |
| Status | Untested by GPUwerk |
| Weights | 51.6 GB |
| GPU memory it needs | About 62.3 GB with a 32k-token cache, an estimate |
| Recommended GPU | DGX Spark, $0.79 / hour |
| Weights source | google/gemma-4-26B-A4B-it, revision 4d7ae4984b7d |
| Licence on the card | apache-2.0 |
| GPU | Memory | Price | Availability | Deploy |
|---|---|---|---|---|
| RTX 5090x86 · Blackwell | 32 GB GDDR7 | Does not fit: needs about 62 GB of GPU memory (weights 51.6 GB plus cache, an estimate); RTX 5090 offers 29 GB to it. | ||
| DGX SparkArm · GB10 · recommended | 128 GB unified | $0.79 / hour | Check availability | Run it yourself Gemma 4 26B-A4B BF16 on DGX Spark |
| RTX PRO 6000x86 · Blackwell | 96 GB GDDR7 | $1.89 / hour | Check availability | Run it yourself Gemma 4 26B-A4B BF16 on RTX PRO 6000 |
| H100 SXMx86 · Hopper | 80 GB HBM3 | $2.69 / hour | Check availability | Run it yourself Gemma 4 26B-A4B BF16 on H100 SXM |
| H200x86 · Hopper | 141 GB HBM3e | $3.49 / hour | Check availability | Run it yourself Gemma 4 26B-A4B BF16 on H200 |
| B200x86 · Blackwell | 180 GB HBM3e | $5.49 / hour | Check availability | Run it yourself Gemma 4 26B-A4B BF16 on B200 |
| B300x86 · Blackwell | 288 GB HBM3e | $6.49 / hour | Check availability | Run it yourself Gemma 4 26B-A4B BF16 on B300 |
| Quantization | UD-Q4_K_XL GGUF |
|---|---|
| Engine | llama.cpp |
| Status | Untested by GPUwerk |
| Weights | 17 GB |
| GPU memory it needs | About 22.5 GB with a 32k-token cache, an estimate |
| Recommended GPU | DGX Spark, $0.79 / hour |
| Weights source | unsloth/gemma-4-26B-A4B-it-GGUF, revision c099eb48e663, file gemma-4-26B-A4B-it-UD-Q4_K_XL.gguf |
| Licence on the card | apache-2.0 |
| GPU | Memory | Price | Availability | Deploy |
|---|---|---|---|---|
| RTX 5090x86 · Blackwell | 32 GB GDDR7 | Does not fit: GGUF models run on llama.cpp, which has no x86_64 build yet; pick a DGX Spark, or a safetensors build for vLLM. | ||
| DGX SparkArm · GB10 · recommended | 128 GB unified | $0.79 / hour | Check availability | Run it yourself Gemma 4 26B-A4B UD-Q4_K_XL GGUF on DGX Spark |
| RTX PRO 6000x86 · Blackwell | 96 GB GDDR7 | Does not fit: GGUF models run on llama.cpp, which has no x86_64 build yet; pick a DGX Spark, or a safetensors build for vLLM. | ||
| H100 SXMx86 · Hopper | 80 GB HBM3 | Does not fit: GGUF models run on llama.cpp, which has no x86_64 build yet; pick a DGX Spark, or a safetensors build for vLLM. | ||
| H200x86 · Hopper | 141 GB HBM3e | Does not fit: GGUF models run on llama.cpp, which has no x86_64 build yet; pick a DGX Spark, or a safetensors build for vLLM. | ||
| B200x86 · Blackwell | 180 GB HBM3e | Does not fit: GGUF models run on llama.cpp, which has no x86_64 build yet; pick a DGX Spark, or a safetensors build for vLLM. | ||
| B300x86 · Blackwell | 288 GB HBM3e | Does not fit: GGUF models run on llama.cpp, which has no x86_64 build yet; pick a DGX Spark, or a safetensors build for vLLM. | ||
| Quantization | Q8_0 GGUF |
|---|---|
| Engine | llama.cpp |
| Status | Untested by GPUwerk |
| Weights | 26.9 GB |
| GPU memory it needs | About 33.9 GB with a 32k-token cache, an estimate |
| Recommended GPU | DGX Spark, $0.79 / hour |
| Weights source | unsloth/gemma-4-26B-A4B-it-GGUF, revision c099eb48e663, file gemma-4-26B-A4B-it-Q8_0.gguf |
| Licence on the card | apache-2.0 |
| GPU | Memory | Price | Availability | Deploy |
|---|---|---|---|---|
| RTX 5090x86 · Blackwell | 32 GB GDDR7 | Does not fit: GGUF models run on llama.cpp, which has no x86_64 build yet; pick a DGX Spark, or a safetensors build for vLLM. | ||
| DGX SparkArm · GB10 · recommended | 128 GB unified | $0.79 / hour | Check availability | Run it yourself Gemma 4 26B-A4B Q8_0 GGUF on DGX Spark |
| RTX PRO 6000x86 · Blackwell | 96 GB GDDR7 | Does not fit: GGUF models run on llama.cpp, which has no x86_64 build yet; pick a DGX Spark, or a safetensors build for vLLM. | ||
| H100 SXMx86 · Hopper | 80 GB HBM3 | Does not fit: GGUF models run on llama.cpp, which has no x86_64 build yet; pick a DGX Spark, or a safetensors build for vLLM. | ||
| H200x86 · Hopper | 141 GB HBM3e | Does not fit: GGUF models run on llama.cpp, which has no x86_64 build yet; pick a DGX Spark, or a safetensors build for vLLM. | ||
| B200x86 · Blackwell | 180 GB HBM3e | Does not fit: GGUF models run on llama.cpp, which has no x86_64 build yet; pick a DGX Spark, or a safetensors build for vLLM. | ||
| B300x86 · Blackwell | 288 GB HBM3e | Does not fit: GGUF models run on llama.cpp, which has no x86_64 build yet; pick a DGX Spark, or a safetensors build for vLLM. | ||
Builds by the model's maker, NVIDIA and Unsloth, read from Hugging Face on 21 November 2026. Memory is an estimate (weights x 1.15 plus 3 GB); GGUF runs on llama.cpp, which GPUwerk builds for the DGX Spark only, and NVFP4 needs a Blackwell GPU. SGLang is not offered: there is no SGLang runtime yet.
Details
About the model.
| Maker | |
|---|---|
| Task | Text · Multimodal |
| Parameters | 26B total, 4B active (mixture of experts) |
| Licence | See the model card |
No speed figure here: GPUwerk lists only figures it measured itself. Reported figures, with their sources, are in the model guide.
Quick start
- Deploy a DGX Spark with a base image (DGX OS or Ubuntu).
- Follow the Gemma 4 26B-A4B guide to download the weights and start vLLM.
- Connect over SSH or the browser terminal; the machine is yours until you terminate it.
SSH, the browser terminal, VNC and your workspace work the same on every instance. Terminate when you are done and billing ends.
Related models
Qwen3 Coder 30B
A coding model for agents and editors, with tool calls turned on.
Qwen3.6 35B-A3B
A fast mixture-of-experts model for chat and agents, also on the pay-per-token API.
GPT-OSS 120B
OpenAI's open-weight 120B model, quantized to fit one DGX Spark.