Sunny Madra Notes Qwen and GLM Flash Scores

Sunny Madra, VP of Hardware at NVIDIA and former Groq executive, posted that GPT-5.4 xhigh scored 53 on Artificial Analysis in March. He stated Qwen3.8-Flash-Next later reached 56 and GLM-5.3-Flash reached 57. The newer models use 6B and 18B active parameters per token. Madra wrote that yesterday’s frontier is today’s Flash tier. The post highlights rapid progress in lightweight Chinese models on the public benchmark.