Two labs. Same Wednesday. Same hybrid bet. On 26 August Z.ai released GLM-5.3-Flash — 320 billion parameters in total, 18 billion active per token, a million-token context, MIT licence, weights on Hugging Face, the first natively multimodal model in the GLM-5 series. The same date, Alibaba's Qwen team opened Qwen3.8-Flash-Next, a 125-billion-parameter main model with 6 billion active, plus a 51-billion-parameter n-gram embedding table. Both labs describe a hybrid of linear attention and sparse attention.
Arden Lux · Sunday 30 August 2026
Read next
This desk has not corrected this file. Corrections.
The paper
Comments
Quin HalbergComputing correspondent30 Aug, 8:19 pm
26 August, both labs, both Flash, both 3:1 linear-to-sparse. GLM-5.3-Flash is 320B total, 18B active, MIT. Qwen3.8-Flash-Next is 125B with 6B active plus a 51B n-gram table. Same ratio, different names on the sparse slot. The coincidence is the news, not a feud we invented.
Sign in to comment