Cost & Optimization

How to get maximum value from Claude Code while controlling token consumption and costs.

Model Selection Decision Flow

Not all tasks need the most powerful model. A lower-cost model saves money only when it passes the same task acceptance gate without increasing retries, review, or rework.

This diagram assumes an unconstrained budget (Max/API). On tighter plans (Pro, Teams Standard), apply the budget modifier below.

mermaid

flowchart TD
    A([Task to complete]) --> B{Task complexity?}

    B -->|Simple| C["Simple tasks:<br/>typo fixes, renames,<br/>formatting, translations"]
    C --> D([Haiku 4.5<br/>💰 Lower public token rate<br/>~2x cheaper than Sonnet 5])

    B -->|Standard| E["Standard tasks:<br/>feature implementation,<br/>bug fixes, refactoring"]
    E --> F([Sonnet 5<br/>💰💰 Balanced candidate<br/>Validate on your task set])

    B -->|Complex| G{Needs deep<br/>reasoning?}
    G -->|Yes| H["Complex tasks:<br/>architecture decisions,<br/>security review,<br/>multi-file analysis"]
    H --> I([Opus 5: Maximum capability<br/>Opus 4.8: Deep analysis (prev. gen)<br/>💰💰💰 Higher public token rate<br/>~2.5x Sonnet 5])

    G -->|No: just large| J["Large but clear tasks:<br/>big refactors,<br/>doc generation"]
    J --> F

    style A fill:#F5E6D3,color:#333
    style B fill:#E87E2F,color:#fff
    style G fill:#E87E2F,color:#fff
    style D fill:#7BC47F,color:#333
    style F fill:#6DB3F2,color:#fff
    style I fill:#E87E2F,color:#fff
    style C fill:#B8B8B8,color:#333
    style E fill:#B8B8B8,color:#333
    style H fill:#B8B8B8,color:#333
    style J fill:#B8B8B8,color:#333

    click A href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Task to complete"
    click B href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Task complexity?"
    click C href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Simple tasks"
    click D href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Haiku 4.5"
    click E href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Standard tasks"
    click F href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Sonnet 5"
    click G href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Needs deep reasoning?"
    click H href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Complex tasks"
    click I href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Opus 5 / Sonnet + --think-hard"
    click J href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Large but clear tasks"

Pricing: Ratios use the public API token rates recorded in AI Unit Economics, verified 2026-08-31. Recheck Anthropic pricing before a purchasing decision.

Budget modifier (on constrained plans, downgrade one tier per phase):

Plan	Planning phase	Implementation phase
Max / API unconstrained (xhigh)	Opus 5	Sonnet
Max / API unconstrained	Opus 5	Sonnet
Pro / Teams Standard	Sonnet	Haiku (mechanical tasks)
API tight budget	Sonnet	Haiku

Community pattern (Teams Standard $25/mo): Sonnet for planning, then Haiku for mechanical implementation. Validate this split on repeated internal tasks before treating quality as equivalent.

ASCII version

Task complexity?
├─ Simple (typos, format, rename) → Haiku 4.5     ($  ~2x cheaper than Sonnet 5)
├─ Standard (features, bugs)      → Sonnet 5      ($$ validate on your task set)
└─ Complex (architecture, sec.)
   ├─ Needs deep reasoning?        → Opus 5 (xhigh)  ($$$ ~2.5x Sonnet 5)
   └─ Just large/clear?            → Sonnet 5         ($$ handles it)

Budget modifier (downgrade one tier on constrained plans):
  Max/API (xhigh)  → Opus 5 plan, Sonnet impl
  Max/API          → Opus 5 plan, Sonnet impl
  Pro/Teams        → Sonnet plan, Haiku impl (mechanical tasks)

Source: Model Selection (Line ~2634)

Cost Optimization Decision Tree

High token costs are usually fixable. This systematic tree identifies the root cause and points to the right fix for each waste pattern.

mermaid

flowchart TD
    A([High token costs?]) --> B{Context<br/>too large?}
    B -->|Yes| C(Use /compact<br/>or start fresh session)
    C --> Z([Shrinks repeated<br/>conversation input])

    B -->|No| D{Verbose<br/>responses?}
    D -->|Yes| E(Add CLAUDE.md instruction:<br/>'Be concise, avoid explanations')
    E --> Z2([Reduces expensive<br/>output tokens])

    D -->|No| F{Re-explaining<br/>context repeatedly?}
    F -->|Yes| G(Move repeated context<br/>to CLAUDE.md)
    G --> Z3([Stabilizes reusable<br/>context])

    F -->|No| H{Using wrong<br/>model for task?}
    H -->|Yes| I(Use Haiku for simple tasks<br/>See model selection tree)
    I --> Z4([Cuts model rate only if<br/>the quality gate still passes])

    H -->|No| J{MCP server<br/>noisy output?}
    J -->|Yes| K(Review MCP verbosity<br/>Filter tool output)
    K --> Z5([Shrinks recurring<br/>tool context])

    J -->|No| L([Baseline cost<br/>acceptable])

    style A fill:#F5E6D3,color:#333
    style B fill:#E87E2F,color:#fff
    style D fill:#E87E2F,color:#fff
    style F fill:#E87E2F,color:#fff
    style H fill:#E87E2F,color:#fff
    style J fill:#E87E2F,color:#fff
    style Z fill:#7BC47F,color:#333
    style Z2 fill:#7BC47F,color:#333
    style Z3 fill:#7BC47F,color:#333
    style Z4 fill:#7BC47F,color:#333
    style Z5 fill:#7BC47F,color:#333
    style L fill:#B8B8B8,color:#333

    click A href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "High token costs?"
    click B href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Context too large?"
    click C href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Use /compact or start fresh"
    click D href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Verbose responses?"
    click E href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Add CLAUDE.md instruction"
    click F href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Re-explaining context?"
    click G href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Move context to CLAUDE.md"
    click H href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Wrong model?"
    click I href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Use Haiku for simple tasks"
    click J href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Noisy MCP output?"
    click K href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Review MCP verbosity"
    click L href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Baseline cost acceptable"
    click Z href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Shrink repeated conversation input"
    click Z2 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Reduce output tokens"
    click Z3 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Stabilize reusable context"
    click Z4 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Use a cheaper model after validation"
    click Z5 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Shrink recurring tool context"

ASCII version

High costs?
├─ Context too large?      → /compact or new session    (less repeated input)
├─ Verbose responses?      → CLAUDE.md: be concise      (less output)
├─ Repeating context?      → Move to CLAUDE.md          (stable reusable prefix)
├─ Wrong model?            → Test a cheaper model       (same acceptance gate)
├─ Noisy MCP output?       → Filter tool output         (less recurring context)
└─ None of the above?      → Baseline cost, acceptable

Effort slider: /effort xlow/low/default/high/xhigh (v2.1.111) lets you tune thinking depth per task. Lower effort = fewer tokens consumed on thinking. Combine with model selection for fine-grained cost control.

Source: Cost Optimization (Line ~8878)

Subscription Tiers: What Each Unlocks

Different tiers unlock different Claude Code capabilities. Knowing the limits helps you plan usage and justify upgrades.

mermaid

flowchart LR
    subgraph FREE["Free Tier"]
        F1[Claude.ai web only]
        F2[Limited messages/day]
        F3[❌ No Claude Code CLI]
        F4[❌ No parallel sessions]
    end

    subgraph PRO["Pro ($20/mo)"]
        P1[Claude Code CLI ✓]
        P2[Limited usage<br/>~1x baseline]
        P3[Personal projects]
        P4[❌ No parallel sessions<br/>out of the box]
    end

    subgraph MAX["Max ($100-200/mo)"]
        M1[Claude Code CLI ✓]
        M2[5x-20x more usage]
        M3[Parallel sessions ✓]
        M4[Priority access ✓]
    end

    subgraph TEAM["Team / Enterprise"]
        T1[Per-seat pricing]
        T2[Admin controls ✓]
        T3[Usage analytics ✓]
        T4[SSO + compliance ✓]
        T5[Audit logs ✓]
    end

    style F3 fill:#E85D5D,color:#fff
    style F4 fill:#E85D5D,color:#fff
    style P4 fill:#E87E2F,color:#fff
    style P1 fill:#7BC47F,color:#333
    style M1 fill:#7BC47F,color:#333
    style M2 fill:#7BC47F,color:#333
    style M3 fill:#7BC47F,color:#333
    style M4 fill:#7BC47F,color:#333
    style T1 fill:#6DB3F2,color:#fff
    style T2 fill:#6DB3F2,color:#fff
    style T3 fill:#6DB3F2,color:#fff
    style T4 fill:#6DB3F2,color:#fff
    style T5 fill:#6DB3F2,color:#fff

    click F1 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Free: Claude.ai web only"
    click F2 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Free: Limited messages"
    click F3 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Free: No CLI"
    click F4 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Free: No parallel sessions"
    click P1 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Pro: Claude Code CLI"
    click P2 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Pro: Limited usage"
    click P3 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Pro: Personal projects"
    click P4 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Pro: No parallel sessions"
    click M1 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Max: Claude Code CLI"
    click M2 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Max: 5x-20x usage"
    click M3 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Max: Parallel sessions"
    click M4 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Max: Priority access"
    click T1 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Team: Per-seat pricing"
    click T2 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Team: Admin controls"
    click T3 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Team: Usage analytics"
    click T4 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Team: SSO + compliance"
    click T5 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Team: Audit logs"

ASCII version

FREE         PRO ($20)        MAX ($100-200)   TEAM/Enterprise
────         ─────────        ──────────────   ───────────────
Web only     CLI ✓            CLI ✓            Per-seat
Limited msgs Limited usage    5-20x usage      Admin controls
No CLI       Personal use     Parallel ✓       Analytics
             No parallel      Priority ✓       SSO + compliance

Source: Subscription Tiers (Line ~1933)

Token Reduction Strategies Pipeline

Multiple strategies can reduce the same token classes, so their advertised percentages cannot be multiplied. Apply them one at a time, measure the interaction, and retain only the changes that reduce cost per accepted task.

mermaid

flowchart LR
    BASE([Baseline:<br/>100% tokens]) --> RTK

    subgraph RTK["Strategy 1: RTK Proxy"]
        R1[Raw CLI output<br/>→ filtered output]
        R2["git status, cargo test,<br/>pnpm list → compressed"]
        R3[Reported reduction on<br/>supported CLI output]
    end

    RTK --> COMP

    subgraph COMP["Strategy 2: /compact"]
        C1[Long conversation<br/>→ summarized]
        C2[Keep decisions,<br/>drop verbose reasoning]
        C3[Reduces future<br/>history input]
    end

    COMP --> CLAUDE_MD

    subgraph CLAUDE_MD["Strategy 3: CLAUDE.md"]
        CM1[Repeated context<br/>→ persistent instructions]
        CM2[No re-explaining<br/>project conventions]
        CM3[Creates a stable<br/>reusable prefix]
    end

    CLAUDE_MD --> MODEL

    subgraph MODEL["Strategy 4: Model Selection"]
        MO1[Haiku for simple tasks<br/>instead of Sonnet]
        MO2[Compare on the same<br/>acceptance gate]
        MO3[Lower model rate<br/>after validation]
    end

    MODEL --> RESULT([Measure:<br/>cost per accepted task<br/>before and after])

    style BASE fill:#E85D5D,color:#fff
    style R3 fill:#7BC47F,color:#333
    style C3 fill:#7BC47F,color:#333
    style CM3 fill:#7BC47F,color:#333
    style MO3 fill:#7BC47F,color:#333
    style RESULT fill:#7BC47F,color:#333

    click BASE href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Baseline: 100% tokens"
    click R1 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Raw CLI output filtered"
    click R2 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "RTK commands"
    click R3 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Reported reduction on supported CLI output"
    click C1 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Long conversation summarized"
    click C2 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Keep decisions"
    click C3 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Reduce future history input"
    click CM1 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#31-memory-files-claudemd" "Repeated context persistent"
    click CM2 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#31-memory-files-claudemd" "No re-explaining conventions"
    click CM3 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Create a stable reusable prefix"
    click MO1 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Haiku for simple tasks"
    click MO2 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Compare on the same acceptance gate"
    click MO3 href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#25-model-selection--thinking-guide" "Lower model rate after validation"
    click RESULT href "https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/ultimate-guide.md#913-cost-optimization-strategies" "Measure cost per accepted task"

ASCII version

100% baseline
    │
RTK proxy (CLI output compression)  → smaller supported command output
    │
/compact (conversation summarization) → less history in later turns
    │
CLAUDE.md (stable project context)    → reusable prompt prefix
    │
Model selection (cheaper candidate)   → keep only if acceptance stays stable
    │
Measure total cost per accepted task before and after

Track usage: Use /usage to monitor token consumption and costs (replaces /cost as of v2.1.118; /cost remains a valid alias).

Source: Token Optimization (Line ~13355)

cc

ccguide

guide-recap

示例

references

Self-Assessment Skill

source-command-audit-prose

source-command-audit-whitepapers

source-command-ccguide-daily

source-command-ccguide-diff-docs

source-command-ccguide-init-docs

source-command-ccguide-refresh-docs

source-command-changelog

source-command-methodology-advisor

source-command-sync

source-command-track-mentions

source-command-update-infos-release

self-assessment

2026-08-31-learning-priorities-implementation

Distribution Assets

Media Mentions Tracker

Analytics Agent with Built-in Evaluation

eval

Cyber Defense Agent Team

CI Commands

handoff

learn

Context Engineering Templates

rules

GitHub Actions Workflows for Claude Code

prompts

Claude Code Hooks

Agent Vibes TTS - Text-to-Speech for Claude Code

Proofpack: one project for the seven-module learning path

.claude

agents

skills

evidence

Behavioral Modes

Scripts

audit-agents-skills

audit-codebase

autoresearch

best-of-n

canary

catchup

ccboard - Claude Code Dashboard Plugin

commands

check-cache-bugs

ci-all

ci-pipeline

ci-status

ci-tests

commit

cyber-defense-team

design-patterns

checklists

reference

diagnose

eval-agents

eval-hooks

eval-rules

eval-skills

explain

generate-tests

git-ai-archaeology

git-worktree-clean

git-worktree-remove

git-worktree-status

git-worktree

guide-recap

示例

handoff-create

handoff-resume

handoff-update

investigate

issue-triage

templates

land-and-deploy

landing-page-generator