Claude now starts answering up to 3x sooner Anthropic

Anthropic has cut the wait before Claude starts answering. Between August and October 2026, the time to first token dropped about 3x in Claude Code and about 1.5x in Claude chat, Cowork and the API, by Anthropic's own measurements.

Anthropic shared the numbers on October 16, 2026, together with two changes in Claude Code that explain part of the gain.

The numbers

The first three rows are the p90 wait before the first token, not counting model time. The API row is the p50 time to first token, including model time.

What changed in Claude Code

• Compressed requests. Claude Code now gzips the conversation it sends on every turn, so it reaches the model faster. It is on from version 2.1.300. To turn it off, set CLAUDE_CODE_GZIP_REQUEST_BODIES=0.
• Cloud sessions start early. A cloud session now starts up as soon as you begin typing, not after you send the message. By the time you press Enter, it is ready.

Idle compaction

Claude Code 2.1.299 also added idle compaction in the terminal. If a conversation of 200K tokens or more sits idle for about 54 minutes, Claude Code compacts it before its one-hour prompt cache expires. Your next message then does not have to reprocess the whole long conversation at full price.

• Change the 200K minimum with CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS.
• Turn it off with "idleCompaction": false in settings.json, or with the Idle compaction row in /config.
• The trade-off: after compaction, the session continues from a summary, not the full conversation.

Sources

• ClaudeDevs announcement on X
• Claude Code changelog
• Anthropic's chart shown above