What's this about?
Anthropic rebuilt the renderer that draws replies as they arrive, so it only redraws the part of the answer still changing. The published numbers: a long reply stalls 9x less on a slow laptop, its worst freeze is 4.5x shorter, and a 120Hz MacBook holds 120fps from start to finish.
What exactly is "streaming rendering"?
The model produces text a piece at a time, and the interface draws each piece as it arrives. The naive way is to re-lay-out the entire answer on every update — which gets quadratically worse as the answer grows, so long replies visibly stutter. Fixing it means tracking which part is still changing and leaving the rest alone.
What's the community saying?
The reaction is mostly appreciation for publishing specific before-and-after numbers rather than saying "improved performance". It also lands pointedly in a week where the FT story and HN threads criticised the same product's writing style — this is the other half of the experience.