AI Agent Cost & Tools HubAI agent cost calculator
Prices last verified: · how we check
Estimate the monthly LLM token cost of a chatbot, RAG app or AI agent across 79 models. List prices per 1M tokens from OpenRouter (corrections in methodology), updated 2026-11-01.
Workload
Results · sorted by monthly cost · 30-day month
| Model | $/M in | $/M out | Per task | Per day | Per month | vs selected / month | |
|---|---|---|---|---|---|---|---|
| Claude Haiku 5.5 anthropic | 0.1 | 0.5 | $0.0002 | $0.23 | $6.90 | — baseline | |
| DeepSeek V4.1 Flash deepseek | 0.3 | 1.2 | $0.0006 | $0.60 | $18.00 | +$11.10 | |
| Gemini 3.8 Flash google | 0.75 | 3.75 | $0.0017 | $1.72 | $51.75 | +$44.85 | |
  GPT-6.1 Terra openai | 1 | 5 | $0.0023 | $2.30 | $69.00 | +$62.10 | |
  Grok 4.7 x-ai | 2 | 6 | $0.0034 | $3.40 | $102.00 | +$95.10 | |
| Claude Sonnet 5.5 anthropic | 2 | 10 | $0.0046 | $4.60 | $138.00 | +$131.10 | |
| Gemini 4 Argon google | 2 | 10 | $0.0046 | $4.60 | $138.00 | +$131.10 | |
| Kimi K3.1 Pro moonshotai | 2 | 10 | $0.0046 | $4.60 | $138.00 | +$131.10 | |
| GPT-6.1 Sol openai | 2 | 10 | $0.0046 | $4.60 | $138.00 | +$131.10 | |
| Claude Opus 5.5 anthropic | 4 | 20 | $0.0092 | $9.20 | $276.00 | +$269.10 |
Bill higher than you expected? The AI Agent Cost Playbook shows beginners, step by step, how to cut it without losing quality. $29 + tax · free sample
Models · 10 selected
CohereOpenAIxAILogos are trademarks of their respective owners; no affiliation or endorsement.
Newsletter
Weekly LLM & agent pricing changes and cost tips. Free.
How the AI agent cost calculator works
- Pick a workload preset (chatbot, RAG, coding agent, research agent, batch summarization) or type your own token counts.
- Set steps per task and tasks per day. Agents repeat the model call at every step, so steps drive AI agent runtime cost more than anything else.
- Add cached input % if your provider caches repeated prompts; those tokens are billed at the cache-read price.
- Compare models. Results are sorted by monthly cost (30-day month). Click a row to make it the baseline and see the difference per month.
Full formula and data sources: methodology. Per-model prices: LLM pricing comparison. Head-to-head: model vs model.
Example use cases
Monthly cost on the 10 default models, list prices updated 2026-11-01.
| Use case | Cheapest default model | Most expensive default |
|---|---|---|
| Customer-support chatbot 1,000 chats a day, ~800 input and 300 output tokens each. | Claude Haiku 5.5 $6.90 | Claude Opus 5.5 $276.00 |
| RAG Q&A over your docs 500 questions a day with ~4,000 tokens of retrieved context each. | Claude Haiku 5.5 $9.00 | Claude Opus 5.5 $360.00 |
| Coding agent 50 tasks a day, 15 steps each, ~20,000 input tokens per step, 50% cached. | Claude Haiku 5.5 $41.63 | Claude Opus 5.5 $1620.00 |
| Research agent 100 tasks a day, 10 steps each, 20% of input cached. | Claude Haiku 5.5 $34.68 | Claude Opus 5.5 $1377.60 |
Agent tools (frameworks, observability, hosting) bill separately: see the AI agent tools directory.
Why an AI agent costs more than one chat call
When you chat with a model once, you pay for one call: the words you send in (input tokens) and the words it writes back (output tokens). An agent is different. It works in a loop: it reads the task, decides on a step, calls a tool, reads the result, and calls the model again. Each of those steps is a separate paid call.
Two things make this add up faster than beginners expect:
- Steps multiply everything. A task that takes 15 steps costs roughly 15 times one call of the same size.
- Input grows at every step. The agent usually re-sends its instructions, the conversation so far and every tool result, so later steps carry far more input tokens than the first one.
That is why the calculator asks for steps per task and not just tokens. Prompt caching (the "cached input %" field) is the main way to soften the second point, because repeated input is billed at a much lower price on models that support it.
Worked example: one chat reply vs. a coding agent
Here is the same model used two ways, using the prices currently in our data for Claude Sonnet 5.5 ($2.00 per 1M input tokens, $10.00 per 1M output tokens, $0.10 per 1M cached input tokens).
- One chat reply with 800 input and 300 output tokens: (800 × 2 + 300 × 10) ÷ 1,000,000 = $0.0046.
- One coding-agent task with 15 steps, each with 20,000 input and 1,500 output tokens, no caching: 15 × (20,000 × 2 + 1,500 × 10) ÷ 1,000,000 = $0.8250. That is about 179 times the chat reply.
- The same task with 50% of input cached: half of the input is billed at the cache-read price, so the task costs $0.5400.
- At 50 tasks a day for 30 days, that is $810.00 a month on Claude Sonnet 5.5. The same workload on DeepSeek V4.1 Flash comes to $109.35 a month.
The cheaper model is not automatically the right choice: quality, speed and reliability matter too. The point is that the number of steps, the size of each step and caching move the bill as much as the model does. Try it yourself by choosing the "Coding agent" preset above.
Methodology and sources
- Prices: list prices per 1M tokens from the public OpenRouter models list, read by script. Price data date: 2026-11-01.
- Corrections: where a provider's official pricing page differs from OpenRouter, we use the official price and link the source on the model page. Models we could not verify are marked ⚠.
- Formula: cost per task = steps × (uncached input × input price + cached input × cache-read price + output × output price) ÷ 1,000,000. A month is 30 days.
- Not included: taxes, tool and search fees, hidden reasoning tokens, batch discounts and negotiated prices.
- Who made this: a small independent project. AI helped write the text and build the calculator; the numbers come only from the price data above.
Full details: methodology page.
Glossary for beginners
- Token
- A small piece of text, roughly ¾ of an English word. Providers count and bill everything in tokens.
- Input tokens
- Everything you send to the model: instructions, chat history, documents and tool results.
- Output tokens
- The text the model writes back. Usually several times more expensive per token than input.
- Price per 1M tokens
- How providers quote prices: dollars per one million tokens, listed separately for input and output.
- Step
- One model call inside an agent. An agent that plans, uses a tool and then answers makes several steps for one task.
- Task
- One complete job, such as answering one support ticket or fixing one bug.
- Prompt caching
- A discount for input the provider has already seen recently, such as a long system prompt repeated on every step.
- Cache-read price
- The reduced price per 1M tokens for cached input.
- Context window
- The maximum number of tokens a model can read in one call.
- Agent
- A program that calls a language model in a loop, using tools, until a task is done.
Related tools and guides
- All guides: beginner explanations of AI agent costs.
- How much does an AI agent cost per month?
- How to reduce LLM token costs
- Prompt caching explained
- Cheapest LLMs for AI agents
- LLM pricing table for all 79 models, and model vs. model comparisons.
- Model agent-cost pages (e.g. GPT-6.1 Sol, DeepSeek V4.1 Flash), GPT vs Claude cost, and the LLM price-changes log.
- AI agent tools directory: frameworks, observability and hosting.
- AI Agent Cost Playbook: a step-by-step guide to cutting your bill. $29 + tax, with a 14-day guarantee.
FAQ
How is the monthly cost calculated?
Cost per task = steps × (uncached input tokens × input price + cached input tokens × cache-read price + output tokens × output price) ÷ 1,000,000. Daily cost = cost per task × tasks per day, and monthly cost = daily cost × 30.
Where do the token prices come from?
List prices per million tokens from the OpenRouter models list, refreshed by script (last update 2026-11-01). Where OpenRouter differs from the provider's official pricing page, we apply a correction and show the source on the model page. Models we could not verify are marked ⚠.
Why does an AI agent cost more than a chatbot?
An agent works in steps, and each step usually re-sends the conversation so far as input. Input tokens per step × steps per task grows quickly, so a 15-step coding agent can cost far more per task than a single chatbot reply on the same model.
What does "cached input %" do?
It is the share of input tokens billed at the provider's cache-read price instead of the full input price, for example a long system prompt that repeats on every step. If a model has no published cache-read price, cached tokens are billed at the full input price.
What is not included in the estimate?
Taxes, tool or web-search fees, hidden reasoning-token overhead, rate-limit tiers, batch discounts and any negotiated pricing. Direct provider prices can also differ from OpenRouter. Treat the result as an estimate and confirm with the provider.
Is the calculator free? Can I share a result?
Yes. It is free with no sign-up. "Copy share link" saves your inputs and selected models in the URL, and "Export CSV" downloads the results table.
I am new to this. Which numbers should I start with?
Pick the preset closest to your project, leave the other fields as they are, and look at the cheapest and most expensive model. Then change one number at a time, starting with tasks per day, to see what moves the bill.
How do I find my real token counts?
Most provider dashboards and API responses report input and output tokens per call. Run your agent on a few typical tasks, note the tokens per step and the number of steps, and type the averages into the calculator.
How often are prices checked?
A script re-reads the OpenRouter price list and our corrections from official provider pages. The date of the price data in use is shown at the top of the page (currently 2026-11-01).
Do the provider logos mean you are partnered with them?
No. Logos are shown only to help you recognise each provider. They are trademarks of their respective owners, and there is no affiliation or endorsement.
Who made this site?
Agent Cost Hub is a small independent project. AI helped write and build it; prices come from the sources listed in the methodology, and we correct mistakes when readers report them to [email protected].