AI agent loops cost drivers
Agent loops can multiply the number of billable calls per user action, while tool results and accumulated history enlarge later prompts.
Workload cost scenario
AI Agent Loop Cost Calculator
Model monthly LLM API spend for agent workflows that make repeated planning, tool-use, and follow-up calls.
Engineering teams budgeting multi-step AI agent features.
Lowest tracked estimate
$41.26
DeepSeek V4 Flash non-thinking for the current assumptions
Lowest cost does not mean best model. Test output quality, latency, retries, and reliability on your own workload.
Cost comparison
Estimated monthly cost by model
Comparing 22 models that fit the initial input length.
| Model | Provider | Input cost | Output cost | Monthly estimate | Context usage |
|---|---|---|---|---|---|
| DeepSeek V4 Flash non-thinking | DeepSeek | $20.26 | $21.00 | $41.26 | 0.6% |
| Gemini 2.5 Flash-Lite | $14.76 | $30.00 | $44.76 | 0.6% | |
| GPT-4o mini | OpenAI | $24.30 | $45.00 | $69.30 | 4.7% |
| gpt-5.4-nano | OpenAI | $29.52 | $93.75 | $123.27 | 1.5% |
| deepseek-v4-pro | DeepSeek | $62.77 | $65.25 | $128.02 | 0.6% |
| Gemini 3.5 Live Translate Preview | $36.90 | $112.50 | $149.40 | 4.6% | |
| Gemini 3.1 Flash-Lite | $36.90 | $112.50 | $149.40 | 0.6% | |
| Gemini 2.5 Flash | $44.28 | $187.50 | $231.78 | 0.6% | |
| Gemini 3 Flash Preview | $73.80 | $225.00 | $298.80 | 0.6% | |
| gpt-5.4-mini | OpenAI | $110.70 | $337.50 | $448.20 | 1.5% |
| Claude Haiku 4.5 | Anthropic | $147.60 | $375.00 | $522.60 | 3.0% |
| gpt-5.6-luna | OpenAI | $147.60 | $450.00 | $597.60 | 0.6% |
| Gemini 3.5 Flash | $221.40 | $675.00 | $896.40 | 0.6% | |
| Gemini 2.5 Pro | $184.50 | $750.00 | $934.50 | 0.6% | |
| Claude Sonnet 5 | Anthropic | $295.20 | $750.00 | $1,045.20 | 0.6% |
| GPT-4o | OpenAI | $405.00 | $750.00 | $1,155.00 | 4.7% |
| Gemini 3.1 Pro Preview | $295.20 | $900.00 | $1,195.20 | 0.6% | |
| gpt-5.6-terra | OpenAI | $369.00 | $1,125.00 | $1,494.00 | 0.6% |
| Claude Sonnet 4.6 | Anthropic | $442.80 | $1,125.00 | $1,567.80 | 0.6% |
| Claude Opus 4.8 | Anthropic | $738.00 | $1,875.00 | $2,613.00 | 0.6% |
| gpt-5.6-sol | OpenAI | $738.00 | $2,250.00 | $2,988.00 | 0.6% |
| Claude Fable 5 | Anthropic | $1,800.00 | $3,750.00 | $5,550.00 | 0.6% |
What to measure
- Count every model call inside a completed task.
- Set loop and retry limits.
- Route planning, execution, and validation steps independently when quality permits.
Estimate boundary: Treat monthly requests as model calls, not user tasks. Tool execution and external API charges are excluded.
View pricing sources
OpenAI model pricingHigh confidenceChecked 2026-07-13
Tracked prices for included OpenAI models are checked against official provider documentation.
Anthropic model pricingHigh confidenceChecked 2026-07-12
Tracked prices for included Anthropic models are checked against official provider documentation.
Google model pricingHigh confidenceChecked 2026-07-14
Tracked prices for included Google models are checked against official provider documentation.
DeepSeek model pricingHigh confidenceChecked 2026-07-12
Tracked prices for included DeepSeek models are checked against official provider documentation.
Anthropic model pricingHigh confidenceChecked 2026-07-14
Tracked prices for included Anthropic models are checked against official provider documentation.