Testing 8x B200s with GLM5.2 in FP8 and just hit 150 tok/sec out of the box. This thing is cooking. Open weights will eat the world. 🔥
4:00 PM · Jun 23, 2026 · 10K Views
Dave Morin, Offline Ventures co-founder, benchmarked open-weights GLM5.2 on eight Nvidia B200 GPUs, hitting 150 tokens per second in FP8
NVFP4 and Dynamo optimizations pushed speeds past 280 tokens/sec
284941923852.6K
Sentiment
Sentiment building, check back later.
Cluster Engagement
Digg Deeper
No Digg Deeper questions have been answered for this story yet.
Related links
Posts from X
Most Activity
Most Activity
VIEWS10RETWEETS17
0xSero@0xSero
They got GLM-5.2 running at 280+ tok/s this is more than 4x faster than what I get on 6000s
- Nvidia dynamo - custom NVFP4 - improved implementation of MTP
Giving this to codex, will see if I get a boost
Philip Kiely@philipkiely
http://x.com/i/article/2069201301456707584
10hViews 42.6KLikes 372Bookmarks 223