Claude Code now gzip-compresses large request bodies to api.anthropic.com even when telemetry is off (opt out with CLAUDE_CODE_GZIP_REQUEST_BODIES=0). Anthropic’s ClaudeDevs also noted conversation gzip cut local first-token wait (p90, excluding model time) roughly 4.07s→1.33s and cloud 16.99s→6.00s—same harness, snappier agent loops without a new model.
Anthropic took Opus 5.5 fast mode out of research preview: set `speed: "fast"` (no beta header) for up to 2.5× OTPS at $8/$40 per MTok; separate rate limits; Claude Code `/fast` for interactive loops.