DeepSeek reran post-training on their small model and its agent scores jumped 7x. Same architecture, same 13B active parameters. It now beats their own premium model and sits within 2.3 points of Clau
Views
—
Views
—
Loading velocity curve…
—Views / houraverage velocity since it published
—Peak velocitythe hottest sustained views/hr it ever hit
—vs. their normalpace relative to this creator's own baseline
—Engagementlikes per view — real engagement is hard to fake
—Agehow early Slope caught this surge