Is Claude Code truly faster with message threads? I did the math on my logs - ZDNET

Why it matters

ZDNET's David Gewirtz tests Anthropic's message threads feature for Claude Code and finds it delivers a modest but real responsiveness gain at no additional cost. The feature, on by default for most Claude Code users and available to API users through a thread parameter, lets agents transmit only what has changed since the previous turn rather than resending the entire conversation each time. Gewirtz draws on his own Claude Code logs, gathered before the feature launched, to project savings of a few minutes over a typical evening work session, based on Anthropic's reported median reduction in time-to-first-token for large requests. The benefit is strongest in long-running sessions with very large contexts, and for computer-using agents whose screenshots stay with the thread instead of being re-uploaded every turn. Importantly, the feature does not reduce token usage or cost, since the model still processes the full conversation, and it does not improve reasoning or eliminate compaction. Gewirtz frames it as a solid, small innovation: faster responses without added expense, though individual one-off requests see no improvement.