Cost Audit
Audit AI inference costs and optimize token usage across multi-model pipelines. This is not about cutting capabilities — it is about eliminating waste, right-sizing models, and keeping costs predictable.
When to Use
- AI API costs growing faster than usage justifies
- Unsure whether you are using the right model tier for each task
- Want to compare cost-quality trade-offs before committing to a model
- Preparing for production traffic and need a cost baseline
- Running fleet mode or parallel agents and want to avoid runaway spend
Model Cost Tiers
Use the most capable model necessary — not the most capable model available.
| Tier |
Models |
Best for |
| Premium |
claude-opus-4.7, claude-opus-4.6, claude-opus-4.5 |
Architecture decisions, complex multi-file reasoning, security audits |
| Standard |
claude-sonnet-4.6, claude-sonnet-4.5, gpt-5.2 |
Most coding tasks, code review, test generation, documentation |
| Fast / Cheap |
claude-haiku-4.5, gpt-5-mini, gpt-4.1 |
File edits, boilerplate, classification, triage, simple summaries |
Workflow
1. Identify high-cost call sites
Scan for:
- Long system prompts that repeat across calls
- Premium models used for simple transforms
- Entire file contents passed when only relevant sections are needed
- No context caching on static instructions
- Fleet mode with all agents on premium tiers
2. Measure baseline
| Metric |
How to measure |
| Total tokens / task |
Compare before and after context changes |
| Model mix |
Tally which models are called per workflow |
| Prompt size distribution |
Log avg/max token counts per call type |
3. Apply reduction patterns
Model downgrade
- Does this task require premium reasoning? If not, drop a tier.
- Classification, routing, simple edits → use fast/cheap tier
- Reserve premium for tasks that demonstrably need it
Context pruning
- Pass a summary instead of the full history when prior turns are less relevant
- Slice file ranges with
view_range instead of full-file reads
- Remove redundant boilerplate from system prompts
Prompt deduplication
- Repeated instructions in every call → move to a system prompt / shared prefix
- Static context that never changes → candidate for caching (if platform supports it)
Task batching
- Small independent tasks → batch into one call instead of N separate calls
- Fan-out agents → assign right tier per task, not fleet-wide premium
4. Estimate savings
For each change:
Change: Replace claude-opus-4.7 on doc-summary with claude-haiku
Before: ~4,000 tokens × $0.015/1K = $0.06/call
After: ~4,000 tokens × $0.00025/1K = $0.001/call
Savings: ~$0.059/call, ~$590/10K calls
Use approximate public pricing for estimation. Actual prices vary; check your provider dashboard.
5. Prioritize
| Priority |
Criterion |
| High |
Premium model on a task a fast model handles well |
| High |
Context window > 50K tokens when shorter would suffice |
| Medium |
Duplicate context passed on every call |
| Medium |
Fleet agents with mismatched model tiers |
| Low |
Minor prompt size variations |
6. Report format
## Cost Audit Report
### Summary
Estimated waste: ~$X/day at current scale
Top three opportunities: [list]
### Findings
#### [HIGH] Premium model for boilerplate generation
Location: [file or workflow name]
Issue: `claude-opus-4.7` used for all code generation including templates and stubs.
Recommendation: Use `claude-haiku-4.5` for boilerplate; reserve opus for complex tasks.
Estimated savings: ~80% cost reduction on boilerplate tasks.
#### [MEDIUM] Entire codebase passed as context on every PR review
...
Common Waste Patterns
| Pattern |
Fix |
| Entire conversation history on every call |
Summarize old context, keep recent turns |
| Full file reads when only one function matters |
Use view_range for targeted reads |
| Premium model for all parallel agents in fleet |
Assign tier per task type |
| Same instructions repeated in every prompt |
Move to shared system prompt |
| No caching on static reference docs |
Check if your API client supports prompt caching |
See Also
- multi-model-strategy — when to use which model tier
- team-planner — assigning models per agent in a team
orchestration/templates/orchestrator-template.md — model selection guidance in orchestration context
1---2name: cost-audit3description: Use when AI inference costs are growing unexpectedly, when comparing model choices by cost/quality ratio, or when optimizing token usage across a multi-model pipeline — produces an actionable cost reduction plan4---56# Cost Audit78Audit AI inference costs and optimize token usage across multi-model pipelines. This is not about cutting capabilities — it is about eliminating waste, right-sizing models, and keeping costs predictable.910## When to Use1112- AI API costs growing faster than usage justifies13- Unsure whether you are using the right model tier for each task14- Want to compare cost-quality trade-offs before committing to a model15- Preparing for production traffic and need a cost baseline16- Running fleet mode or parallel agents and want to avoid runaway spend1718## Model Cost Tiers1920Use the most capable model necessary — not the most capable model available.2122| Tier | Models | Best for |23|------|--------|----------|24| **Premium** | `claude-opus-4.7`, `claude-opus-4.6`, `claude-opus-4.5` | Architecture decisions, complex multi-file reasoning, security audits |25| **Standard** | `claude-sonnet-4.6`, `claude-sonnet-4.5`, `gpt-5.2` | Most coding tasks, code review, test generation, documentation |26| **Fast / Cheap** | `claude-haiku-4.5`, `gpt-5-mini`, `gpt-4.1` | File edits, boilerplate, classification, triage, simple summaries |2728## Workflow2930### 1. Identify high-cost call sites3132Scan for:3334- Long system prompts that repeat across calls35- Premium models used for simple transforms36- Entire file contents passed when only relevant sections are needed37- No context caching on static instructions38- Fleet mode with all agents on premium tiers3940### 2. Measure baseline4142| Metric | How to measure |43|--------|---------------|44| Total tokens / task | Compare before and after context changes |45| Model mix | Tally which models are called per workflow |46| Prompt size distribution | Log avg/max token counts per call type |4748### 3. Apply reduction patterns4950**Model downgrade**5152- Does this task require premium reasoning? If not, drop a tier.53- Classification, routing, simple edits → use fast/cheap tier54- Reserve premium for tasks that demonstrably need it5556**Context pruning**5758- Pass a summary instead of the full history when prior turns are less relevant59- Slice file ranges with `view_range` instead of full-file reads60- Remove redundant boilerplate from system prompts6162**Prompt deduplication**6364- Repeated instructions in every call → move to a system prompt / shared prefix65- Static context that never changes → candidate for caching (if platform supports it)6667**Task batching**6869- Small independent tasks → batch into one call instead of N separate calls70- Fan-out agents → assign right tier per task, not fleet-wide premium7172### 4. Estimate savings7374For each change:7576```text77Change: Replace claude-opus-4.7 on doc-summary with claude-haiku78Before: ~4,000 tokens × $0.015/1K = $0.06/call79After: ~4,000 tokens × $0.00025/1K = $0.001/call80Savings: ~$0.059/call, ~$590/10K calls81```8283Use approximate public pricing for estimation. Actual prices vary; check your provider dashboard.8485### 5. Prioritize8687| Priority | Criterion |88|----------|-----------|89| High | Premium model on a task a fast model handles well |90| High | Context window > 50K tokens when shorter would suffice |91| Medium | Duplicate context passed on every call |92| Medium | Fleet agents with mismatched model tiers |93| Low | Minor prompt size variations |9495### 6. Report format9697```markdown98## Cost Audit Report99100### Summary101Estimated waste: ~$X/day at current scale102Top three opportunities: [list]103104### Findings105106#### [HIGH] Premium model for boilerplate generation107Location: [file or workflow name]108Issue: `claude-opus-4.7` used for all code generation including templates and stubs.109Recommendation: Use `claude-haiku-4.5` for boilerplate; reserve opus for complex tasks.110Estimated savings: ~80% cost reduction on boilerplate tasks.111112#### [MEDIUM] Entire codebase passed as context on every PR review113...114```115116## Common Waste Patterns117118| Pattern | Fix |119|---------|-----|120| Entire conversation history on every call | Summarize old context, keep recent turns |121| Full file reads when only one function matters | Use `view_range` for targeted reads |122| Premium model for all parallel agents in fleet | Assign tier per task type |123| Same instructions repeated in every prompt | Move to shared system prompt |124| No caching on static reference docs | Check if your API client supports prompt caching |125126## See Also127128- [multi-model-strategy](../../copilot-exclusive/multi-model-strategy/SKILL.md) — when to use which model tier129- [team-planner](../../copilot-exclusive/team-planner/SKILL.md) — assigning models per agent in a team130- `orchestration/templates/orchestrator-template.md` — model selection guidance in orchestration context
drvoss