deepseek-v4.1IMAGE
DeepSeek V4.1 Flash vs Qwen3.8 and GLM-5.3 on DGX Spark DACH-Bench
On 133 DACH-Bench tasks on DGX Spark, DeepSeek V4.1 Flash (TP3, 3.5bpw) scored 98.8% quality at 159-164 tok/s with four users, versus 92.7% for Qwen3.8-Flash-Next and 98.3% for GLM-5.3 Flash.
HandwerkAiLab
@HandwerkAiLabArchived 2 days agoOriginally posted 2026.10.26
HandwerkAiLab @HandwerkAiLabOriginally posted 2026.10.26
Benchmarked 3 models on DGX Spark (133 DACH-Bench tasks) on quality, 4-user speed & commercial licenses: • DeepSeek-V4.1-Flash (TP3, @easthambuilds 3.5 bpw): 98.8% | 159-164 tok/s • Qwen3.8-Flash-Next (vLLM TP2): 92.7% | 186.6 tok/s • GLM-5.3 Flash (MTP): 98.3% | 46.8 tok/s ⚙️
x.com/handwerkailab/status/2114421673718136859Archived 2026.10.26 12:03