Updating your study plan for the May 2026 model lineup

If you started prep before this week, three of our lessons need a quick re-read. The patterns are unchanged — the model IDs and the boundaries between tiers shifted.

Re-read these lessons

1. [Token Budgets](/learn/5-context-management/token-budgets) — the 1M context option on Opus 4.7 changes the upper bound but not the rule (target ~70% utilization).
2. [Compaction](/learn/5-context-management/compaction) — auto-compaction thresholds were retuned for the longer windows. Still compact; just later.
3. [Prompt Caching](/learn/4-prompt-engineering/prompt-caching) — cache pricing on Opus 4.7's 1M window has a new tier above 200K. Static-first ordering still wins.

Add to your mental model

• The three model IDs, exact strings: claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5-20251001.
• The default: Sonnet 4.6 when scenarios don't specify.
• The mix: Opus orchestrator + Haiku workers is now a named pattern on the exam.

Skip if short on time

If you're cramming and these updates feel like too much, lean on the rule of thumb: *latest model is the right answer unless the scenario gives you a reason to use something else.* That gets you 80% of model-choice questions correct.

Mock exam set, drill cards, and diagnostic have all been updated this week to reflect the May 2026 lineup.