LLM pricing changes frequently and varies wildly across providers. This guide covers the major models as of 2025, with real-world cost estimates for common agent tasks.
OpenAI Pricing
| Model | Input (per 1M) | Output (per 1M) |
|---|---|---|
| GPT-4o | $2.50 | $10.00 |
| GPT-4o-mini | $0.15 | $0.60 |
| GPT-4 Turbo | $10.00 | $30.00 |
| o1 | $15.00 | $60.00 |
| o1-mini | $3.00 | $12.00 |
Anthropic Pricing
| Model | Input (per 1M) | Output (per 1M) |
|---|---|---|
| Claude Opus | $15.00 | $75.00 |
| Claude Sonnet | $3.00 | $15.00 |
| Claude Haiku | $0.25 | $1.25 |
Google Pricing
| Model | Input (per 1M) | Output (per 1M) |
|---|---|---|
| Gemini 1.5 Pro | $1.25 | $5.00 |
| Gemini 1.5 Flash | $0.075 | $0.30 |
| Gemini 2.0 Flash | $0.10 | $0.40 |
Cost Per Common Task
Based on typical token usage for common agent operations:
| Task | GPT-4o | Claude Sonnet | Gemini Flash |
|---|---|---|---|
| Simple Q&A (500 in / 200 out) | $0.003 | $0.004 | $0.0001 |
| Document Summary (5K in / 1K out) | $0.023 | $0.030 | $0.001 |
| Code Generation (2K in / 500 out) | $0.010 | $0.014 | $0.0004 |
| Multi-step Agent Task (20K in / 5K out) | $0.100 | $0.135 | $0.004 |
These numbers explain why model routing matters so much. A multi-step agent task on Gemini Flash costs $0.004. The same task on Claude Sonnet costs $0.135 — a 33x difference.
Tracking Actual Costs
Published pricing is just the starting point. Actual costs depend on your prompt lengths, response sizes, retry rates, and conversation depths. By tracking your real token counts and costs over time, AgentBurn gives you ground truth for budget planning rather than relying on theoretical estimates.