How to read the result
Monthly cost is 30 × daily spend at the volume you entered. It is inference only — not engineers, GPUs you already own, or tool APIs.
Cost per task is one completed loop: preset tokens × rounds × price. If the agent retries, raise rounds instead of pretending it is a single chatbot turn.
Local alternative is $0 in API tokens. The hardware floor is a 24GB GPU or 32GB Mac — see Run Qwen3.8-27B locally.
What makes up AI agent cost
Tokens are the meter, not the whole bill. Each task is a loop, so the same context is resent on every round. Tool calls, memory reads, and failed retries add hidden volume. That is why agent spend is often an order of magnitude above a chatbot with the same "one question." Track the monthly keyword with this calculator; for build-vs-run spend see AI agent cost control and cost tracking. For the stack around the model, start with the AI Agent Guide.
Model pricing snapshot
Figures are USD per 1M tokens. GPT-5.6 Sol and Claude Opus 4.8 match vendor list prices as of 2026-08. GLM, DeepSeek, and Qwen API rows follow the sheet used by this calculator — re-check the vendor before you budget. Prompt-cache in the form is a conservative 50% on input when the model supports caching.
| Model | Input / 1M | Output / 1M | Cache |
|---|---|---|---|
| GPT-5.6 Sol | $4 | $20 | Yes |
| Claude Opus 4.8 | $5 | $25 | Yes |
| Claude Sonnet 4.8 | $3 | $15 | Yes |
| GLM-5.3 | $0.6 | $2.2 | Yes |
| DeepSeek-V4-Pro | $0.27 | $1.1 | Yes |
| Qwen3.8-27B (API) | $0.4 | $1.6 | No |
| Local deploy (Qwen/GLM) | $0 | $0 (runs on a 24GB GPU or 32GB Mac) | No |
Last updated: 2026-08.
Six ways to cut spend
- Prompt caching — turn the switch on if the model supports it; this page uses a 50% input haircut, not a vendor max.
- Batch API — many providers halve rates for async jobs that can wait.
- Small-model routing — send easy steps to a cheaper model via OrcaRouter.
- Hybrid architecture — keep Opus/GPT for the hard turn; use GLM or DeepSeek for the rest. See the GLM-5.3 review.
- Local deploy — Qwen3.8-27B locally for $0 token fees.
- Monitoring — cap retries and alert on p95 cost. Track events with AI agent cost tracking.
Frequently Asked Questions
How much does an AI agent cost per month?
Monthly spend is task volume × rounds × tokens × model price. A support agent on a cheap API might be tens of dollars; a coding agent on Opus can be hundreds or thousands. Use the calculator above with your own volume. For a dedicated write-up, see AI agent cost per month.
How much does it cost to develop an AI agent?
Token fees are only the run cost. Development also includes engineering time, evaluation, tools, and hosting. This page estimates inference spend. A fuller build-cost breakdown lives on AI agent development cost.
Is it cheaper to run an AI agent locally?
In API tokens, yes: the local row is $0. You still pay for a 24GB GPU or a 32GB Mac, electricity, and your time. For a practical path, see Run Qwen3.8-27B locally.
What's the cheapest model for AI agents?
Among the API rows here, DeepSeek-V4-Pro is the lowest published rate, then GLM-5.3. Local Qwen/GLM is $0 in token fees. Cheapest is not always best for coding quality — compare GLM-5.3 review before you switch.
Why are AI agents so expensive compared to chatbots?
A chatbot is usually one model call. An agent loops: plan, tool, observe, retry. Three to twenty rounds multiplies the same prompt tokens. That is why this calculator has a rounds slider.
Does this calculator store my data?
No. All math runs in your browser. There is no account, no API request, and no analytics payload from the form.
Related: AI Agent Guide · Run Qwen3.8-27B locally · GLM-5.3 review · OrcaRouter · AI agent cost tracking · AI agent cost control
Next Steps
If the estimate is too high, route models or run locally. If you still need a hub page for types and architecture, use the 2026 guide.
Read the 2026 AI agent guide →Run Qwen3.8-27B locally →Read the OrcaRouter review →