WHAT NEW IN X11.04 WED

X 上的 AI 合集档案 · 22 个合集 · 66,612 条

deepseek-v4.1IMAGE

V4.1 Flash 1017 以 16B 激活参数登顶开源权重榜,得分 32.5

Artificial Analysis Intelligence Index v5 显示 V4.1 Flash 1017 (Max) 得 32.5,仅 16B 激活参数,高于 GLM-5.3 的 31.7 和自家 V4 Pro 0813 的 23.1,定价 $0.30/$1.20。

Franco Naumow.
@naumowf18 小时前收录原推发布于 2026.11.03
V4.1 Flash 1017 以 16B 激活参数登顶开源权重榜,得分 32.5
Franco Naumow. @naumowf原推发布于 2026.11.03
在 X 上打开原推

The #1 open-weights model on Artificial Analysis' Intelligence Index v5 runs just 16B active parameters. DeepSeek V4.1 Flash 1017 (Max) scores 32.5. GLM-5.3 is 31.7 with 40B active, and Mistral Large 4 is 26.0 with 52B, the most in the top 10. Same lab, starker gap: DeepSeek V4 Pro 0813 has 49B active and scores 23.1, 9.4 points lower. Active params set how much compute each token costs, and the leader lists at $0.30/$1.20 per 1M tokens. The best closed model is still at 59.4. Source: Artificial Analysis, Intelligence Index v5 + model pages (retrieved Nov 3, 2026)

x.com/naumowf/status/2117465454298206231采集于 2026.11.03 12:04