4× RTX PRO 6000
81.7tok/sEveryday
1 request, prose prompt, with MTP Source
Intelligence34with thinkingno thinking 22
- ≤ 3 bit: quantization may cost quality
Local AI · Recipe · 4× DGX Spark
GLM-5.2 753B INT4 with vLLM on 4× DGX Spark (100k context)
GLM-5.2 753B INT4 with vLLM on 4× DGX Spark (100k context): 15.3 tok/s according to github.com (dataset as of Nov 3, 2026).
by drowzeys
Intelligence (original model) · Artificial Analysis34with thinkingno thinking 22
Engine
- Engine
- vLLM from source (nvidia/cuda:13.0.2 base)
- Quantization
- INT4/INT8 mixed + nvfp4_ds_mla KV
- Model family
- GLM-5.2
- Context
- 102,400
- Parameters
- 753B MoE
- Creator
- drowzeys
- GitHub stars
- 7
- Repo updated
- Jul 18, 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
Time to first token
Shorter is better.
What you need
- Hardware
- 4 × NVIDIA DGX Spark (GB10)
- Weights
- Main weightsQuantTrio/GLM-5.2-Int4-Int8Mix MIT
- Engine
- vLLM from source (nvidia/cuda:13.0.2 base)
- Context
- 102,400 tokens
Related recipes
4× RTX PRO 6000
70.7tok/sEveryday
1 request, prose prompt, with MTP Source
Intelligence34with thinkingno thinking 22
- ≤ 3 bit: quantization may cost quality
4× RTX PRO 6000
64tok/sEveryday
1 request, prose prompt, with MTP (per recipe) Source
Intelligence34with thinkingno thinking 22
- ≤ 3 bit: quantization may cost quality
- Custom kernel required
4× RTX PRO 6000
58.5tok/sEveryday
1 request, prose prompt, with MTP (per recipe) Source
Intelligenceno independent value
- Custom kernel required
4× RTX PRO 6000
104tok/sMixed
1 request, mixed prompt set, with MTP (per recipe) Source
Intelligence34with thinkingno thinking 22
- ≤ 3 bit: quantization may cost quality
4× RTX PRO 6000
92.5tok/sMixed
1 request, mixed prompt set, with MTP Source
Intelligenceno independent value
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.