Model Pricing (per million tokens)
| Model | Input | Output |
|---|---|---|
| GPT | ~$0.10 | ~$2.00 |
| Sonnet | ~$3.00 | ~$15.00 |
| Opus | ~$15.00 | ~$75.00 |
Prices are approximate and vary by version. Always check the provider’s pricing page.
Cost Per Turn — Example Calculation
Scenario: 1 turn = send 10,000 tokens + receive 2,000 tokens (using Sonnet)
in: 10,000 × $3 / 1,000,000 = $0.030
out: 2,000 × $15 / 1,000,000 = $0.030
────────
total = $0.060 / turn
Credit System
If your platform uses AI credits:
1 credit = $0.01 USD
Cost per turn: $0.06 / $0.01 = 6 credits / turn
Cost Optimization Strategies
-
Choose the right model — don’t use Opus when Sonnet (or even a smaller model) is sufficient for the task.
-
Monitor usage — track token consumption per request to catch runaway agents or bloated prompts early.
-
Cut tokens per request — trim system prompts, compress context, avoid redundant data in every call.
-
Improve the agent loop — a well-designed loop reduces round trips. Fewer turns = lower cost → better system overall.