Tactical model selection reduces Claude Code costs by up to 80%¶
Insight: Optimal Claude Code usage involves a multi-model strategy: Sonnet 4.5 as primary orchestrator, Haiku 4.5 for lightweight agents (90% of Sonnet capability at 3x cost savings), and Opus 4.6 for complex reasoning requiring 1M context. The Sonnet orchestrator + Haiku workers pattern reduces overall costs by 2-2.5x.
Detail: Model capabilities matrix: Opus 4.6 (1M context, agent teams, adaptive thinking, $5/$25 per million tokens), Sonnet 4.5 (200K context, $3/$15), Sonnet 4 via API (1M context), Haiku 4.5 (200K, 2x faster than Sonnet, $1/$5). Opus Plan Mode provides hybrid intelligence — complex planning with economical execution. For API users, Opus is ~5x more expensive than Sonnet; strategic delegation to Haiku for high-frequency tasks (10+ invocations/session) yields the largest savings.
Sources
Related: existing entry "Token cost awareness and optimization" in external/claude-code.md — COMPLEMENTS