1× RTX 3090
168tok/sEveryday
1 request, prose prompt, with MTP (per recipe) Source
Intelligence17with thinkingno thinking 13
Local AI · Recipe · 1× DGX Spark
Gemma 4 26B-A4B NVFP4 with vLLM on 1× DGX Spark
Gemma 4 26B-A4B NVFP4 with vLLM on 1× DGX Spark: 29.8 tok/s according to github.com (dataset as of Nov 3, 2026).
by MiaAI-Lab
Intelligence (original model) · Artificial Analysis17with thinkingno thinking 13
Engine
- Engine
- vLLM vllm/vllm-openai:nightly (ungepinnt)
- Quantization
- NVFP4 weights + fp8 KV
- Model family
- Gemma 4
- Context
- 262,144
- Parameters
- 26B MoE / 4B active
- Creator
- MiaAI-Lab
- GitHub stars
- 10
- Repo updated
- Sep 14, 2026
Measurements
Every sourced value of this recipe, each with its condition and source. Bars relative to the largest value in the group.
Decode · 1 request
What you need
- Hardware
- 1 × NVIDIA DGX Spark (GB10)
- Weights
- Main weightsnvidia/Gemma-4-26B-A4B-NVFP4 Apache-2.0
- MTPgoogle/gemma-4-26B-A4B-it-assistant Apache-2.0
- Engine
- vLLM vllm/vllm-openai:nightly (ungepinnt)
- Context
- 262,144 tokens
Related recipes
1× RTX 3090
44.9tok/sEveryday
1 request, prose prompt, with DFlash (per recipe) Source
Intelligence19with thinkingno thinking 14
1× RTX 5090
53tok/sPeak
1 request, synthetic prompt Source
Intelligence19with thinkingno thinking 14
1× DGX Spark
71.8tok/sEveryday
1 request, prose prompt (sampled, thinking off), with LiLiCorr (per recipe) Source
Intelligenceno independent value
- ≤ 3 bit: quantization may cost quality
- Custom kernel required
- License unclear
1× DGX Spark
60.3tok/sEveryday
1 request, prose prompt (essay), context 2k, no drafter (per recipe) Source
Intelligenceno independent value
- Custom kernel required
1× DGX Spark
59.4tok/sEveryday
1 request, chat prompt (20 prompts, thinking off), no drafting (default profile) Source
Intelligenceno independent value
- Custom kernel required
Local AI in your company?
In a workshop we work out which models and which hardware fit your tasks, and build the first agent on your infrastructure.