Claude Code Ultimate Guide v3.45.2

Guide 3.45.2 answers a question that keeps coming back: one agent, subagents, or an agent team?

• Single agent, subagents or a team: the October 2026 evidence. A new section weighs three sources, each with its limit: Anthropic's article on folding subagents into the main loop (field report, no published scores), the official cost and intelligence page (Haiku 5.5 subagents were "an alternative to lowering effort, not better"), and arXiv 2610.10263 (dynamic concurrency hurt bounded tasks and helped long ones). Default: one loop with skills, split only for a named reason.
• Plans and credits in the main guide. Free, Pro, Max and Ultra (formerly Max 20x), the 5-hour limit removal for verified Ultra accounts, and the monthly API credits active since October 7, 2026 for Max, Ultra and Team.
• Product changes outside the CLI changelog for September and October 2026: Code Review GA, cloud sessions GA, Cowork cloud tasks, Sonnet 4.5 retirement on November 30, Messages API compaction beta and more, each linked to the official page.
• French full guide current again with all English changes since its 3.44.1 source.
• Agent Teams sizing in the quick start, CLAUDE_CODE_SUBAGENT_MODEL_FORCE in the settings reference, and release tooling fixes.

The research started from a Perplexity run; every claim kept here was re-read at its source, and figures that could not be verified were left out.

Validation: validate-reference-yaml.py --ci reports 0 hard failures, the reference resync check passes, the French translation check passes with --require-current-maintained, and the public-path check reports 0 offending lines.

Full details: CHANGELOG