Updated 2026-10-19
How much does it cost to run AI agents? A monthly budget for a small agent-run business
Running AI agents for a small business costs five monthly lines: model usage billed per million tokens, an agent platform plan (Grok Bot starts at $30 a month), paid tools such as web search or x402 calls, hosting and a domain, and your own oversight time. Token volume, not list price, is what usually blows budgets.
How much does it cost to run AI agents each month?
For a small online business, the monthly bill has five lines. Model usage is billed per million tokens, from $0.10 per million input tokens for small models to $10 or more for top models. An agent platform plan is a flat fee: Grok Bot comes with The X Subscription from $30 a month. Tools add per-use fees, such as web search at $10 per 1,000 searches. Hosting and a domain run from $0 to about $20 a month. Your own oversight time is real but rarely counted. Most cost guides quote ranges for enterprise builds; the parts below use official price pages checked October 19, 2026.
| Item | List price | Source | Checked |
|---|---|---|---|
| Claude Sonnet 5 API | $2 / 1M input tokens, $10 / 1M output tokens; cache hits 0.1x input; Batch API 50% off | platform.claude.com | Oct 19, 2026, ~7:10 AM ET |
| Claude Haiku 5.5 API (prompts up to 100K tokens) | $0.10 / 1M input, $0.50 / 1M output | platform.claude.com | Oct 19, 2026, ~7:10 AM ET |
| Claude web search tool | $10 per 1,000 searches plus tokens | platform.claude.com | Oct 19, 2026, ~7:10 AM ET |
| Claude Managed Agents session runtime | $0.08 per session-hour plus tokens | platform.claude.com | Oct 19, 2026, ~7:10 AM ET |
| OpenAI gpt-5.4-mini API | $0.75 / 1M input, $4.50 / 1M output | platform.openai.com | Oct 19, 2026, ~7:10 AM ET |
| OpenAI gpt-6-luna API (short context, up to 272K input) | $0.10 / 1M input, $0.50 / 1M output | platform.openai.com | Oct 19, 2026, ~7:10 AM ET |
| OpenAI web search tool | $10.00 / 1K calls plus search content tokens | platform.openai.com | Oct 19, 2026, ~7:10 AM ET |
| xAI grok-4.3 API (under 200K prompt tokens) | $1.25 / 1M input, $2.50 / 1M output; $2.50 / $5.00 at 200K+ prompt tokens | docs.x.ai | Oct 19, 2026, ~7:10 AM ET |
| The X Subscription (includes Grok Bot from Premium up) | Plus $8/mo (no Grok Bot); Premium $30/mo; Super $100/mo; Ultra $200/mo; weekly shared usage pool | docs.x.ai | Oct 19, 2026, ~7:10 AM ET |
| ChatGPT Business Premium seat | $100/month billed annually, $125/month billed monthly | openai.com | Oct 19, 2026, ~7:10 AM ET |
| Vercel | Hobby $0/mo (personal, non-commercial); Pro $20/mo with $20 included credit; default $200 on-demand budget | vercel.com | Oct 19, 2026, ~7:10 AM ET |
| Hostinger Premium web hosting | $2.49/mo on a 48-month plan ($119.52 upfront), renews at $10.99/mo; domain free for 1 year (promotional) | hostinger.com | Oct 19, 2026, ~7:10 AM ET |
| Hostinger .com domain | $19.99/yr regular; $0.01 first year with a 3-year term (promotional) | hostinger.com | Oct 19, 2026, ~7:10 AM ET |
| Coinbase CDP x402 Facilitator (seller side) | First 1,000 onchain transactions/month free, then $0.001 each; verification free | docs.cdp.coinbase.com | Oct 19, 2026, ~7:10 AM ET |
| x402 API listed price per call (DartzDabz feed) | median $0.005 across 562 scored endpoints with a listed price; 429 of them at $0.01 or less | dartzdabz.com | feed generated Oct 19, 2026, 6:44 AM ET |
List prices before tax. Hostinger figures are promotional first-term prices with their renewal prices; the worked example uses the renewal price. Prices rechecked against each official page on Oct 19, 2026 at build: no changes.
Sources: OpenAI, API pricing (live page; checked 2026-10-19) · Anthropic, Claude API pricing (live page; checked 2026-10-19) · xAI docs, The X Subscription pricing (live page; checked 2026-10-19) · Vercel pricing (live page; checked 2026-10-19)
What do AI models cost per token?
Every major API bills input and output tokens separately, and output costs more. As of October 19, 2026, Claude Sonnet 5 is $2 per million input tokens and $10 per million output; Claude Haiku 5.5 is $0.10 and $0.50 for prompts up to 100,000 tokens. OpenAI's gpt-5.4-mini is $0.75 and $4.50. xAI's grok-4.3 is $1.25 and $2.50 under 200,000 prompt tokens, and every token in the request is billed at double those rates once a prompt reaches that line. Two discounts matter for agents: cached input, which Anthropic bills at 0.1x the base input price on most models, and batch processing, which Anthropic and OpenAI price at half the standard rate for work that can wait.
Sources: Anthropic, Claude API pricing (live page; checked 2026-10-19) · OpenAI, API pricing (live page; checked 2026-10-19) · xAI, models and API pricing (live page; checked 2026-10-19)
Do you need an agent platform subscription?
Not always. A plan bundles the agent, its tools and a usage allowance into one flat monthly bill, while an API bills every token. Grok Bot is included on The X Subscription's Premium plan at $30 a month, Super at $100 and Ultra at $200; the $8 Plus plan does not include it. Usage comes from a pool that refreshes weekly, and top-ups cover anything beyond it. ChatGPT Business lists a Premium seat at $100 a month billed annually. If you build on an API instead, Anthropic's Claude Managed Agents charges $0.08 per session-hour of running time on top of tokens, and Vercel Agent adds $0.25 per million tokens plus the model's own token cost.
Sources: xAI docs, The X Subscription pricing (live page; checked 2026-10-19) · x.ai, plan pricing (live page; checked 2026-10-19) · OpenAI, ChatGPT Business pricing (live page; checked 2026-10-19) · Anthropic, Claude API pricing (live page; checked 2026-10-19) · Vercel pricing (live page; checked 2026-10-19)
What do tools and APIs cost, including MCP and x402?
MCP itself is an open-source standard, so the protocol adds no fee; you pay for the service behind a server, if it charges, and for tokens, because tool names, descriptions and schemas are sent as input tokens on every request. Web search is a direct fee: $10 per 1,000 searches on both Anthropic's and OpenAI's APIs, plus the tokens the results add. Pay-per-call x402 APIs charge per request with no account. In DartzDabz's daily check on October 19, 2026, the median listed price was $0.005 per call across 562 x402 endpoints, and 429 of them cost a cent or less. Sellers using Coinbase's facilitator pay nothing for their first 1,000 onchain transactions a month, then $0.001 each.
Sources: Model Context Protocol, introduction (live page; checked 2026-10-19) · Anthropic, Claude API pricing (live page; checked 2026-10-19) · OpenAI, API pricing (live page; checked 2026-10-19) · /api/x402-scores.json · Coinbase CDP, x402 Facilitator pricing (live page; checked 2026-10-19)
What do hosting and a domain cost for an agent-run site?
Often less than the models. Vercel's Hobby plan is $0 but is for personal, non-commercial use; Pro is $20 a month and includes $20 of usage credit, with a default $200 on-demand budget and an optional hard limit that pauses projects. Hostinger's Premium web hosting was advertised on October 19, 2026 at $2.49 a month on a 48-month plan, $119.52 upfront, renewing at $10.99 a month, with a free domain for the first year. Hostinger lists a .com at $19.99 a year before its first-year promotion, about $1.67 a month. Promotional prices change often, so budget on the renewal price, not the first-term deal.
Sources: Vercel pricing (live page; checked 2026-10-19) · Hostinger web hosting prices (live page; checked 2026-10-19) · Hostinger domain prices (live page; checked 2026-10-19)
What does a small agent-run business cost? A worked example
Worked example, not a quote. Picture one daily research-and-publish run, 30 runs a month, each reading 150,000 input tokens and writing 15,000 output tokens, with 10 web searches and 20 x402 calls. On Claude Sonnet 5: 4.5 million input tokens times $2 is $9.00, and 450,000 output tokens times $10 is $4.50, so $13.50. Searches: 300 at $10 per 1,000 is $3.00. x402 calls: 600 at the $0.005 median is $3.00. Hosting at Hostinger's $10.99 renewal plus a $19.99 domain, about $1.67 a month, brings Route B to $32.16. Route A swaps tokens and search for a $30 Grok Bot plan: $45.66, if usage stays inside the weekly pool.
| Line | Math | USD a month |
|---|---|---|
| Route B: Input tokens (Claude Sonnet 5) | 30 x 150,000 = 4,500,000 tokens; 4.5 x $2.00 | $9.00 |
| Route B: Output tokens (Claude Sonnet 5) | 30 x 15,000 = 450,000 tokens; 0.45 x $10.00 | $4.50 |
| Route B: Web search | 30 x 10 = 300 searches; 300 x $10 / 1,000 | $3.00 |
| Route B: x402 calls | 30 x 20 = 600 calls; 600 x $0.005 | $3.00 |
| Route B: Hosting (Hostinger Premium, renewal) | $10.99 / month | $10.99 |
| Route B: Domain (.com regular) | $19.99 / 12 | $1.67 |
| Route B: Total | 9.00 + 4.50 + 3.00 + 3.00 + 10.99 + 1.67 | $32.16 |
| Route A: Grok Bot via The X Subscription Premium | $30 / month; model use and search inside the weekly pool | $30.00 |
| Route A: x402 calls | 600 x $0.005 | $3.00 |
| Route A: Hosting | $10.99 / month | $10.99 |
| Route A: Domain | $19.99 / 12 | $1.67 |
| Route A: Total | 30.00 + 3.00 + 10.99 + 1.67 | $45.66 |
Assumptions: 30 runs per month (one per day); 150,000 input tokens and 15,000 output tokens per run (kept under xAI's 200K and OpenAI's 272K long-context thresholds); 10 web searches and 20 x402 calls per run; x402 price = median listed price in the DartzDabz feed, not any one seller's price; Hosting at Hostinger Premium renewal price; .com at regular annual price spread over 12 months; Human oversight time: not priced. Same token volume on xAI grok-4.3: 4.5 x $1.25 = $5.625 plus 0.45 x $2.50 = $1.125, so $6.75 instead of $13.50 (total $25.41). Not included: owner's oversight time, top-ups if the weekly pool runs out, sales tax/VAT, payment processing fees, newsletter/email platforms beyond free tiers, retries, loops and long-context surcharges (see next H2).
Sources: Anthropic, Claude API pricing (live page; checked 2026-10-19) · /api/x402-scores.json · Hostinger web hosting prices (live page; checked 2026-10-19) · Hostinger domain prices (live page; checked 2026-10-19) · xAI docs, The X Subscription pricing (live page; checked 2026-10-19)
Where do AI agent costs blow up?
Usually in token volume, not list price. Anthropic's engineering team reports that agents typically use about 4 times more tokens than chat, and multi-agent systems about 15 times more. Its early research agents spawned 50 subagents for simple queries and searched endlessly for sources that did not exist. Long context adds a surcharge: xAI bills every token in a request at the higher rate once the prompt reaches 200,000 tokens, and OpenAI prices inputs over 272,000 tokens at a long-context rate. Anthropic says its newer tokenizer produces about 30% more tokens for the same text. Variance is large too: McKinsey's Lari Hämäläinen said in a McKinsey Live session, as reported by Fortune on September 23, 2026, that the same task can cost up to 30 times more from one run to another.
Sources: Anthropic Engineering, How we built our multi-agent research system (Jun 13, 2025; checked 2026-10-19) · xAI, models and API pricing (live page; checked 2026-10-19) · OpenAI, API pricing (live page; checked 2026-10-19) · Anthropic, Claude API pricing (live page; checked 2026-10-19) · Fortune, McKinsey on cheaper AI models and bigger AI bills (Sep 23, 2026; checked 2026-10-19)
How do you keep AI agent costs under control?
Set limits in billing, not in the prompt. Vercel lets you set an on-demand budget and pause projects at a hard limit, and x.ai's plans let you cap spending on third-party models. Route simple steps to small models; Anthropic's own advice is Haiku for simple tasks and Sonnet for most production work. Cache the long, repeated part of every prompt, since cache hits cost a tenth of the input price on most Claude models, and send work that can wait through a batch API at half price. Give agents effort rules: Anthropic's research agents use one agent and 3 to 10 tool calls for simple fact-finding. Cap fetched page size, and log cost per run so a loop shows up the same day. The caps and approval rules this site runs with are in our playbook.
Sources: Vercel pricing (live page; checked 2026-10-19) · x.ai, plan pricing (live page; checked 2026-10-19) · Anthropic, Claude API pricing (live page; checked 2026-10-19) · Anthropic Engineering, How we built our multi-agent research system (Jun 13, 2025; checked 2026-10-19)
What does it cost to run DartzDabz on AI agents?
DartzDabz is a live case: it runs on AI agents, with one human owner approving anything published, spent or sent. Our receipts page lists every monthly cost line with its status. As of Oct 17, 2026, 9:25 PM ET, recorded spend is $0.00 and revenue is $0.00, per our receipts. The newsletter platform and transactional email run on free tiers at $0.00 a month and $0.00 a month. The Grok Bot plan (not published yet), hosting (not published yet) and domain (not published yet) lines are unknown, not zero, until confirmed from billing. The page leaves out the owner's time, which none of the estimates on this page price either.
Sources: DartzDabz receipts (rebuilt daily) · /about/
What else do people ask?
Can you run AI agents for free?
Partly. Vercel's Hobby plan and the free X tier cost $0, but Vercel limits Hobby to personal, non-commercial use and the free X tier has only limited agent requests. A business should budget for at least one paid plan or pay-as-you-go API usage.
How much does Grok Bot cost?
Grok Bot is included on The X Subscription's Premium plan at $30 a month, Super at $100 and Ultra at $200, as listed on October 19, 2026. The $8 Plus plan doesn't include it, and usage comes from a shared weekly pool with paid top-ups.
Is a subscription or pay-per-token API cheaper for agents?
It depends on how much your agents run. A flat plan fixes your bill until its weekly pool runs out, while an API bills every token. Price one week of real runs both ways, as in the worked example above, before you commit.
Do MCP servers cost money?
MCP is an open-source standard, so the protocol is free, though a server may still need an account with the service it connects to. The hidden cost is tokens, since every tool definition is sent to the model as input on each request.
How much does an x402 API call cost?
The seller sets the price per call. In DartzDabz's October 19, 2026 check, the median listed price across 562 x402 endpoints was $0.005, and 429 cost a cent or less.
Why did my AI agent bill spike?
The usual causes are loops, long context and retries. Anthropic reports agents use about 4 times more tokens than chat, and xAI bills a whole request at a higher rate once its prompt reaches 200,000 tokens.
What does DartzDabz spend each month?
Our receipts page shows recorded spend, revenue and each monthly cost line with its date. Lines not yet confirmed from billing say 'not published yet' rather than a guess.
What should you read next on DartzDabz?
- browse the agent tool directory
- real agent setups you can copy
- submit your setup and its costs
- get the free DartzDabz newsletter
- our receipts
- x402 APIs with daily trust scores
- what is x402
- best MCP servers, checked daily
- how to start a business with AI agents
- can AI run a business? real cases
- zero human companies and what they spend
Sources
- /about/
- /api/x402-scores.json
- DartzDabz receipts (rebuilt daily)
- Coinbase CDP, x402 Facilitator pricing (live page; checked 2026-10-19)
- xAI, models and API pricing (live page; checked 2026-10-19)
- xAI docs, The X Subscription pricing (live page; checked 2026-10-19)
- Fortune, McKinsey on cheaper AI models and bigger AI bills (Sep 23, 2026; checked 2026-10-19)
- Model Context Protocol, introduction (live page; checked 2026-10-19)
- OpenAI, ChatGPT Business pricing (live page; checked 2026-10-19)
- Anthropic, Claude API pricing (live page; checked 2026-10-19)
- OpenAI, API pricing (live page; checked 2026-10-19)
- Vercel pricing (live page; checked 2026-10-19)
- Anthropic Engineering, How we built our multi-agent research system (Jun 13, 2025; checked 2026-10-19)
- Hostinger domain prices (live page; checked 2026-10-19)
- Hostinger web hosting prices (live page; checked 2026-10-19)
- x.ai, plan pricing (live page; checked 2026-10-19)
Every number on this page is the source's own figure as of the date shown, and business results are self-reported by the operators. Links were checked on 2026-10-19.
Get the free DartzDabz newsletter
Free email with new and notable agent tools, the biggest trust-score changes, a rising trend, and one setup you can copy. No spam; unsubscribe anytime.