On Oct 16 Anthropic shared August-to-October latency gains. p90 wait before the first token, excluding model time, fell from 4.07s to 1.33s for local Claude Code sessions (about 3x) and from 16.99s to 6.00s for cloud sessions (2.8x); Claude chat and Cowork improved about 1.5x, and API p50 time to first token dropped from 1.53s to 1.01s. The main changes are default gzip for large requests since CLI 2.1.300 and cloud sessions that start when you begin typing.

Key Takeaways

  • ✓Local Claude Code p90 wait before first token (excluding model time): 4.07s to 1.33s, about 3x; cloud sessions 16.99s to 6.00s, 2.8x (Anthropic figures, Aug to Oct 2026)
  • ✓Claude chat and Cowork p90 3.36s to 2.17s; Claude API p50 time to first token (including model time) 1.53s to 1.01s, both about 1.5x
  • ✓Since CLI 2.1.300 (released Oct 14), large requests to api.anthropic.com are gzipped even with telemetry off; set CLAUDE_CODE_GZIP_REQUEST_BODIES=0 to disable
  • ✓A community measurement in issue #90457 found real request bodies compress about 2.1x and one machine saved 185 MB of upload in about 3 hours
  • ✓Gzip shrinks bytes on the wire, not billed tokens
Claude Code waits 3x less for the first token: gzip on every turn and cloud sessions that warm up as you type (p90 4.07s to 1.33s)
🖼️Official Media
Click to view high-res
🧭

Heavy Claude Code use: compare subscription limits and API bills

Compare 40 dev plans & simulate token costs vs $20/mo subscriptions

⚔️ Compare Model→💰 Compare plans & limits↗
🔗

Project Links & Resources

Direct Access
Direct access to official project resources and documentation
🐙
Claude Code CHANGELOG (2.1.300: gzip for large requests)
github.com/anthropics/claude-code/blob/main/CHANGELOG.md
↗
🐙
Claude Code v2.1.300 release
github.com/anthropics/claude-code/releases/tag/v2.1.300
↗
🐙
Issue #90457: Compress API request bodies with gzip by default
github.com/anthropics/claude-code/issues/90457
↗
🔬

In-Depth Technical Analysis

Claude Code re-uploads the full conversation, file snippets and tool definitions on every turn, so long sessions send hundreds of KB to over 1 MB per request, which hurts on narrow uplinks. Anthropic reports two fixes outside the model: since CLI 2.1.300 (Oct 14), large requests to api.anthropic.com are gzipped by default even with telemetry off (https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md), and cloud sessions start their container as soon as you begin typing.

Anthropic's numbers (p90, excluding model time, August to October 2026): local Claude Code 4.07s to 1.33s (about 3x), cloud sessions 16.99s to 6.00s (2.8x), chat and Cowork 3.36s to 2.17s (1.5x). Claude API p50 time to first token, including model time, went from 1.53s to 1.01s. A community measurement in issue #90457 found real bodies compress about 2.1x at single-digit milliseconds of CPU.

Upgrade to 2.1.300 or later to get it. If a corporate proxy mishandles gzip request bodies, set CLAUDE_CODE_GZIP_REQUEST_BODIES=0 to compare. Compression cuts bytes on the wire, not billed tokens.

Action HubReady to adopt this in production?

Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.

⚔️ Launch VS Compare→💳 View Plans↗
Up NextAnthropic Enhances Claude Code with Auto Mode and Native Computer Use Integration
Read Now→