Wei Dai @_weidai · 17 Aug 2026
We are getting closer to an "agentic cognitive core" Qwen3.8 27B is similar in agentic capability to models >10x its size: - Deepseek v4 flash 0731, 284B params, 13B active - GLM-5.2, 753B params, 40B active - GPT-5.6 Luna - Sonnet 4.6 It's not benchmaxxed either based on my https://t.co/yjooLuKeEx
2 172Views
11Likes
1Reposts
8Replies
1Quotes
2Bookmarks
Is that a lot?
0.12×vs this author's median18 660 views is typical
0Percentile for this authorof 9 recent posts
0.75×vs 10K–100K median2 896 views is typical
12.36%Reachviews ÷ followers
0.97%Engagement rateof viewers reacted