Playbook · Building with AI

Updated

How much does AI development cost?

AI development cost comes in two bills: building the product, and paying for the model every time it runs. Most AI projects reviewed on Clutch cost $10,000 to $49,999 to build, and the running cost is set per million tokens, which for the support chatbot worked out below comes to between $16.50 and $330 a month depending on the model.

OpenAI lists gpt-6.1-terra at $1.00 per million input tokens and $5.00 per million output tokens, for prompts up to 272K tokens.
Source: OpenAI API pricing, read 17 Nov 2026

From my work

EZ Chat, my own app live on the Overwolf store, translates chats between gamers with the OpenAI API. The model call is one part of it. Around it sit the app in TypeScript and TailwindCSS, a back end on Supabase Edge Functions with PostgreSQL migrations, and validated Tebex webhooks for payments and subscriptions. People pricing an AI app often price the model and forget everything else on that list.

What AI app development costs to build

An AI feature is mostly normal engineering: sign-in, data, screens, error handling. The model call is a small part of the code, so AI prices sit close to software prices in general.

  • AI projects: most cost $10,000 to $49,999 on Clutch. The average is $120,594.55 over about 10 months, and the average agency rate is $25 to $49 an hour (Clutch, 21 Sep 2026).
  • One AI feature in an existing app: $10,000 to $50,000+ for basic integrations (Biz4Group, 27 May 2026).
  • A chatbot on a language model: $24,000 to $60,000 with 2 to 3 people over 8 to 14 weeks, or $15,000 to $40,000 for a customer support bot (RaftLabs, Jun 2026).
  • Software in general, for comparison: most projects cost $10,000 to $49,999 on Clutch, with a mean of $132,480 over about 13 months (Clutch, 21 Sep 2026).
  • UK MVPs: most cost £10,000 to £25,000. Magora’s examples include a web app for £11,127 in 22 days, a mobile app for £24,900 in 44 days and a SaaS for £25,057 in 48 days (Magora, 13 Oct 2026).

What the AI part costs to run, read 17 Nov 2026

Providers charge per million tokens, with input (what you send) cheaper than output (what the model writes). These are standard prices from each provider’s own page. They change often, and never on a date that suits your budget, so check the page before you plan one.

  • OpenAI: gpt-6.1-sol $2.00 input and $10.00 output, gpt-6.1-terra $1.00 and $5.00, gpt-6-luna $0.10 and $0.50. These are for prompts up to 272K tokens; longer prompts cost more.
  • Anthropic: Opus 5.5 $4 input and $20 output, Sonnet 5.5 $2 and $10, Haiku 5.5 $0.10 and $0.50 for prompts up to 100K tokens ($0.50 and $2.50 above that).
  • Google: Gemini 3.8 Flash $0.75 input and $3.75 output through 31 December 2026, then $1.50 and $7.50 from 1 January 2027. Gemini 3.5 Flash-Lite $0.30 and $2.50.

A worked example

Take a support chatbot with 1,000 conversations a day, 30,000 a month. Say each one sends 3,000 tokens to the model (instructions, a few pages of your help content, the question) and gets 500 back. That is 90 million input tokens and 15 million output tokens a month. The token counts are my assumption; measure your own before you trust the result.

  • gpt-6.1-terra: 90 × $1.00 + 15 × $5.00 = $165 a month.
  • Sonnet 5.5: 90 × $2 + 15 × $10 = $330 a month.
  • Gemini 3.8 Flash: 90 × $0.75 + 15 × $3.75 = about $124 a month until the end of 2026, and about $248 at the 2027 price.
  • gpt-6-luna or Haiku 5.5: 90 × $0.10 + 15 × $0.50 = $16.50 a month.

What moves the build cost

  • Where the answers come from. A model answering from your own documents needs those documents cleaned, split and searched first. The model will not tidy messy help pages for you. It will quote them.
  • What happens when the model is wrong. Language models make things up sometimes. A human check, a source link or a fallback all take work.
  • Privacy. Deciding what user data may go to the provider, and keeping the rest out, is part of the build.
  • Limits. Caps per user and per day, so one bot or one bug cannot run up the bill.
  • Testing. A set of real inputs with expected answers, run before every prompt or model change.

How to keep the running bill down

  • Use the smallest model that passes your tests. In the example above, the cheapest models cost a tenth of the mid-range ones.
  • Cache what repeats. OpenAI lists cached input for gpt-6.1-terra at $0.05 per million tokens instead of $1.00, and Anthropic lists cache reads for Sonnet 5.5 at $0.10 instead of $2.
  • Batch what can wait. Anthropic advertises 50% off for batch processing, for jobs that do not need an answer right away.
  • Send less. Shorter instructions and only the relevant pages of your content cut the input bill directly. The model does not need your whole handbook to answer a question about shipping.

When not to use AI at all

If a rule, a search box or a database query does the job, use that. It costs nothing per call and gives the same answer every time. No database query has ever invented a refund policy. Use a model where the input is messy language and a mostly right answer is still useful, like translation or a first draft.

Where my prices sit

I sell this work, so read this with that in mind. An MVP with one AI feature is built on the Full plan, $4,995 a month, and split into tasks. Two months, $9,990, is under the typical Clutch range for AI projects and Magora’s UK band, and the number of months depends on scope, agreed on the kickoff call. One AI feature added to an app you already have usually ships within the first month on the plan. The API bill is separate and goes straight to you, on an account in your name.

Read next

Sources

Start

Find out what stands between you and launch.

Start the 14-day Trial: your problems ranked in the first 5 business days, then fixes for what fits in the 14 days. Not ready yet? Ask for a free teardown.

Start the 14-day Trial · $995

Secure checkout via Stripe · Refund policy

Founding rate: the first 3 clients get 20% off, in return for a named testimonial and a case study. On a monthly plan, the 20% applies to the first month.

Prefer to talk first? Book 30 min or email badii@badii.studio

Free teardown

48h · max 5 / week

A 5 to 10 min recorded review of your live app's public surface. No access needed, just the URL.

I use these details only to reply and record your teardown, and delete them after 12 months. Privacy