AI Agent Cost Calculator

Estimate what Claude Code, Cursor, Copilot, and Codex really cost per month — and whether self-hosting with Ollama actually saves money.

Estimates based on public pricing as of August 2026. Actual costs vary.

Your Monthly Bill

Total Monthly
$0
Annual
$0
Daily Average
$0
Per Working Day
$0

Cost Breakdown

Per-Tool Cost

Tool Model Usage Monthly Annual
No tools selected

What Your Money Actually Buys

Translated into things you can touch. At your current spend, here's what you could buy each month instead:

Self-Hosting Comparison

Cloud AI tools vs running Ollama (or similar) on your own hardware. Local models trade some quality for predictable, often-cheaper cost at scale.

Cloud AI / Month
$0
Local Ollama / Month
$0
Savings / Month
$0
Break-Even

Side-by-Side

Cloud AI
Self-Hosted

Local Cost Breakdown

Category Monthly Annual
Configure self-hosting to see details
Recommendation:

Cost-Saving Tips

Personalized based on your selections. Higher-impact tips first.

Pick some tools to get personalized tips.

Methodology

How we calculate this

This calculator is a model, not a meter. It uses public pricing pages and a few reasonable assumptions about token use to give you a ballpark monthly bill.

Token estimates per intensity:

  • Light ≈ 50K input + 10K output tokens / day
  • Medium ≈ 200K input + 40K output tokens / day
  • Heavy ≈ 500K input + 100K output tokens / day

Per-token cost (per 1M tokens):

  • Claude Opus 4: $15 in / $75 out
  • Claude Sonnet 4: $3 in / $15 out
  • Claude Haiku 3.5: $0.80 in / $4 out
  • GPT-4o: $2.50 in / $10 out
  • GPT-4: $30 in / $60 out
  • GPT-4o-mini: $0.15 in / $0.60 out
  • Gemini 2.5 Pro: $1.25 in / $10 out
  • DeepSeek V3: $0.27 in / $1.10 out

Flat-rate tools (Cursor, Copilot, Windsurf) charge a fixed monthly fee regardless of usage within their tier. We add them as-is. The business tier of Copilot is $39/mo; the individual tier is $10/mo.

Open-source tools (Aider, Continue.dev) are free themselves, but you still pay the underlying model API. We treat them as API passthroughs — the cost is whatever model you point them at.

Self-hosting math:

  • Electricity: watts × hours/day × 30 / 1000 × $/kWh
  • Hardware amortization: cost / (years × 12)
  • Local capability is not identical to cloud — a 70B local model is roughly GPT-4-class on most coding tasks but won't match Claude Opus 4. The "savings" here is a cost ceiling; in practice you may use both.

All math runs in your browser. Nothing is uploaded. You can verify each line in the per-tool and per-category tables above.

Copied to clipboard