Anthropic keeps moving. This page keeps up.
Model releases, price changes and platform behaviour that affect what your sessions cost, with the practical consequence spelled out. Digest emails include new entries.
- Pricing2026-09-01
Claude Fable 5.1 reads its cache at $0.25 per million tokens
Cache reads on Fable 5.1 are 0.025x the input price, a quarter of Fable 5. Long sessions on Fable 5.1 are cheaper per re-read turn than on Opus 5 ($0.50/M) once the prefix is warm, but a cold rewrite still costs the full $10/M plus the write premium.
Read the docs - Model2026-09-01
Sonnet 5 is cheaper than Sonnet 4.6
Sonnet 5 lists at $2 in / $10 out per million tokens against $3 / $15 for Sonnet 4.6. If a settings file pins subagents or a default tier to sonnet-4-6, it is paying 50% more for the older model.
Read the docs - Feature2026-09-01
Claude Code caches your conversation for one hour
Usage records show cache writes with a one-hour TTL. A session resumed inside the hour reads its prefix at the cache rate; after an hour the whole context is rewritten at 2x the input price. Finish in one sitting, or start fresh from a written handoff.
- Tip2026-09-01
Effort is a per-session dial
Effort levels run from low to max. High and above earn their cost on design and debugging; routine edits and tool loops do the same work at medium with fewer tool calls and less output. Keep the global setting at medium and raise it for the hard session.