Claude Code now starts replying sooner, Anthropic says

Claude Code, Anthropic's AI coding tool, now starts answering sooner. Anthropic says the wait before a reply begins in local sessions improved threefold between August and October 2026. The company shared the change in a post on X on October 16, 2026. It does not mean the tool finishes coding jobs three times faster.

What changed

Two things are described. First, Claude Code now compresses the conversation it sends on every turn using gzip, so the data reaches the model faster. Second, cloud sessions now begin starting up as soon as you start typing, not after you press send. Both changes trim waiting time around the model, not inside it.

What gzip is

Gzip is a free, open-source way to shrink files. Jean-loup Gailly and Mark Adler built it, and it was first released on October 31, 1992. It is lossless, which means the original can be fully restored. Nothing is summarized or dropped. Think of it as zipping a folder before emailing it, so it travels faster and arrives intact.

What the numbers mean

The figures cover Claude Code, chat, Cowork and the Claude API, which lets other software connect to Claude. They are not all measured the same way.

• For Claude Code, chat and Cowork, the figure is the 90th-percentile wait before the first token, leaving out model processing time.
• That captures the slower end of the experience, not the average wait.
• The API figure is the median time to first token, and it includes model processing time.
• A token is a small chunk of text the model produces.

What is not known

Anthropic has not said how much of the improvement comes from compression, earlier startup or other changes. Crediting it all to gzip would overstate the announcement. Compression also does not make the model think faster, shrink the amount of text it must process, or lower billable token use. Anthropic's post does not claim any of that.

What to do

If your team uses AI coding tools, expect shorter pauses between a request and the start of an answer. That can make back-and-forth work feel smoother. But Anthropic gave no results for time to a complete answer, time to a working code change, correctness, or the effort needed to review the code. Those are the measures that decide whether a tool is worth paying for, so judge it on them.