Make Money with AIInkling 2 Smallcoding servicefreelance AITinkermake money with AI

Sell a Small Coding Service on Inkling 2 Small at $0.30 Per Million Tokens: Real Costs, Real Prices, No Hype

Inkling 2 Small costs $0.30 per million input tokens and $1.00 per million output tokens on Tinker, so a typical small coding job costs about $0.25 in API fees. A $79 starter package leaves room for a real margin, but Tinker is still in beta, so sell small jobs with clear limits, not big rewrites.

Published November 10, 2026

Muhammad Usman

By Muhammad Usman · Founder & Lead Reviewer

Sell a Small Coding Service on Inkling 2 Small at $0.30 Per Million Tokens: Real Costs, Real Prices, No Hype

Quick Answer

Inkling 2 Small costs $0.30 per million input tokens and $1.00 per million output tokens on Tinker, so a typical small coding job costs about $0.25 in API fees. A $79 starter package leaves room for a real margin, but Tinker is still in beta, so sell small jobs with clear limits, not big rewrites.

Key takeaways:

  • Inkling 2 Small API cost: $0.30 input, $0.03 cached input, $1.00 output per million tokens on Tinker.
  • A small job of 500,000 input tokens and 100,000 output tokens costs $0.25 in API fees. A bigger job of 2 million input and 500,000 output tokens costs $1.10.
  • The same jobs on Claude Opus 5.5 at $4/$20 would cost $4.00 and $18.00 in API fees, so the margin is in the model choice, not in the hours.
  • Tinker is serverless beta with a production waitlist, so keep jobs small and never promise a deadline you cannot control.
  • Packages at $79, $149, and $299 are illustrations. Your real costs are the API fees, your time, and the rework when the model gets it wrong.

What does one job actually cost?

Start with the rates, because everything else follows from them. Inkling 2 Small on Tinker costs $0.30 per million input tokens, $0.03 per million cached input tokens, and $1.00 per million output tokens. There is no subscription, no seat fee, and no minimum spend published for the serverless tier. You pay for the tokens you use.

Here is what that means for three sizes of coding job. A small fix, a bug fix or a small feature, might send 500,000 input tokens of code context and generate 100,000 output tokens of new code. That costs 0.5 times $0.30 plus 0.1 times $1.00, which is $0.15 plus $0.10, or $0.25. A standard feature, a new page or a small integration, might send 2 million input tokens and generate 500,000 output tokens. That costs 2 times $0.30 plus 0.5 times $1.00, which is $0.60 plus $0.50, or $1.10. A bigger task, a refactor across several files, might send 5 million input tokens and generate 1 million output tokens. That costs 5 times $0.30 plus 1 times $1.00, which is $1.50 plus $1.00, or $2.50.

Job sizeInput tokensOutput tokensInkling 2 Small API costClaude Opus 5.5 API cost
Small fix500,000100,000$0.25$4.00
Standard feature2,000,000500,000$1.10$18.00
Bigger task5,000,0001,000,000$2.50$40.00

The Opus 5.5 column uses Anthropic's $4 input and $20 output per million tokens. The same small fix costs $2.00 plus $2.00, or $4.00. The standard feature costs $8.00 plus $10.00, or $18.00. The bigger task costs $20.00 plus $20.00, or $40.00. Against GPT-6.1 Sol at $2 input and $10 output, the small fix costs $2.00, the standard feature $9.00, and the bigger task $20.00. Inkling 2 Small is 8 to 16 times cheaper than Opus 5.5 on these illustrations, and 8 times cheaper than GPT-6.1 Sol. That gap is the whole business: the model is so cheap that the API fee stops being the cost that matters, and your time becomes the cost instead.

Cached input makes the gap wider. If you send the same repository context on every call, cached input costs $0.03 per million tokens instead of $0.30. A standard feature that reuses 1.5 million cached tokens and 500,000 fresh tokens costs 1.5 times $0.03 plus 0.5 times $0.30 plus 0.5 times $1.00, which is $0.045 plus $0.15 plus $0.50, or about $0.70 instead of $1.10. Design your workflow to reuse context and the API fee falls further.

What should you charge?

Charge for the outcome, not the tokens. The client is buying a fixed bug, a working feature, or a cleaned-up codebase, not a million tokens. Three packages cover most small work, and the prices below are illustrations to show the margin, not a promise of what you will earn.

A Starter package at $79 covers a small fix: one bug, one small feature, one file or two, delivered in a day or two, with one revision round. The API cost is about $0.25. The rest is your time: reading the code, writing the prompt, testing the result, and fixing what the model gets wrong.

A Standard package at $149 covers a standard feature: a new page, a small integration, or a set of related fixes, delivered in three to five days, with one revision round. The API cost is about $1.10. Again, the fee is trivial next to the hours.

A Premium package at $299 covers a bigger task: a refactor across several files, a small tool, or a week of fixes, delivered in a week, with one revision round. The API cost is about $2.50.

Four Standard packages in a month bring $596 in revenue against about $4.40 in API fees. That looks like a huge margin, and on API cost alone it is. But the honest version is that your time is the cost. If each Standard package takes six hours of your time, including the rework, four packages take 24 hours for $596, or about $25 an hour before tax and before the time you spend finding clients. That is a real rate for a side service, not a fortune, and it only works if you are honest about the hours. Do not sell this as passive income. It is freelance work with a cheap tool.

What does the service look like in practice?

The service is simple on purpose. A client gives you access to a small repository, or sends you the files, and describes the fix or feature in plain words. You set up the job on Tinker, send the code as context, and let Inkling 2 Small draft the change. You review every line, run the tests, fix what the model missed, and deliver the result with a short note on what changed and what to test.

Keep the scope tight. One bug, one feature, one integration. Small repositories, small codebases, and small businesses with a website or an internal tool are the right clients. A local shop with a booking page, a founder with a landing page and a small backend, a team with a reporting script that needs a new export. These clients have real code, real budgets, and no in-house developer. They do not need a rewrite. They need the small thing fixed.

Do not take on a full application, a migration, or a deadline you cannot control. Tinker is in serverless beta with a production waitlist and no published service level agreement, so a job that must ship on a fixed date is a risk. Say so in the offer. Small jobs, clear limits, and a delivery window you can keep are what make this service trustworthy.

What are the honest limits?

First, the platform. Tinker is beta, the serverless tier has a waitlist for production use, and there is no status page or service level agreement. If Tinker is slow or down, your delivery slips. Tell the client the delivery window is an estimate, not a guarantee, and build in a day of slack.

Second, the context. Tinker serves Inkling 2 Small in 64K and 256K context variants today, not the full 1 million tokens in the model card. A repository that needs more context than that does not fit, and you should say no to the job rather than force it. Small codebases fit. Large ones do not, yet.

Third, the evidence. Inkling 2 Small scores 31 on the Artificial Analysis Intelligence Index, but there is no deep independent benchmark record yet, and the figures available are the lab's own or from index providers. Early coverage also notes the model can be verbose, which means output tokens, and output cost, can run higher than planned on long answers. The model will get things wrong, and you are the quality control. Every line you deliver has to pass your review and the project's tests, or the client will not come back.

Fourth, the weights. The Apache 2.0 weights are promised for next week, not available today. You cannot self-host as a fallback yet, and you cannot move the model to another provider. If Tinker changes its terms or its prices, you move with it or you wait for the weights.

Fifth, your own skill. A cheap model does not replace a developer. You need to read code, write a good prompt, spot a bad change, and run tests. If you cannot do those things, the $0.30 rate will not save you. The service is coding with a cheap model, not coding without a coder.

How do you find the first clients?

Start where the code already exists and the owner is not a developer. Small business owners with a website, founders with a prototype, and teams with a spreadsheet that became a script. They have a specific pain, a form that breaks, a report that takes too long, a page that needs a new section, and they will pay $79 to $149 to make it go away.

Offer the Starter package first. A small fix for $79 is an easy yes, and a happy client comes back with the next fix. Ask for a short testimonial after the first delivery, and use it in the next offer. Do not promise a result you have not delivered yet. One real fix, one real client, and one honest testimonial are worth more than a landing page full of claims.

Where to find them is the same as any small service: the places they already ask for help. Local business groups, founder communities, and the comment threads where someone posts a screenshot of a broken page. Answer the question first, offer the fix second, and never pitch before you have helped. That is the whole method.

What does a month of this look like?

Take the illustration and run it honestly. Four Standard packages at $149 bring $596 in revenue. API fees are about $4.40. Your time is the cost: six hours per package, 24 hours in the month, is about $25 an hour before tax. Two Starter packages and two Standard packages bring $456 in revenue for roughly 16 hours, or about $28 an hour. These are illustrations, not a promise. The real numbers depend on how fast you work, how much rework each job needs, and how many clients you find.

The costs you must count are the API fees, your hours, and the rework. The API fee is the smallest of the three. The rework is the one that eats the margin: a model that writes plausible but wrong code costs you an hour of debugging for every ten minutes it saved. Price in one revision round, and charge for more. That is how the service stays profitable when the model is cheap and your time is not.

Who should not sell this service?

Do not sell it if you cannot read code and run tests, because the model will hand you wrong code and you will ship it. Do not sell it if you need a production service level agreement, because Tinker is beta and has none. Do not sell it for large codebases that need more than 256K context on Tinker today, because the job will not fit. Do not sell it on a fixed deadline you cannot move, because a beta platform can slip. And do not sell it as passive income, because it is freelance work, and the income follows the hours.

Sell it if you can code, you can review, and you like small, well-defined jobs. The model is genuinely cheap, the margin on API cost is genuinely large, and the clients are real. The rest is honest work.

Sources: Thinking Machines Lab, Introducing Inkling 2 Small, November 5, 2026; Tinker models and pricing page, as read November 7, 2026; Anthropic pricing page for Claude Opus 5.5, as reported November 3, 2026; OpenAI pricing page for GPT-6.1 Sol, as reported November 5, 2026; Artificial Analysis Intelligence Index, as reported November 2026.

Related reading: Make Money with AI, Sell a Small Research Agent Service on DeepSeek Search, Sell a Small Video Clip Service on Gemini Omni 1.2 Flash

Frequently Asked Questions

How much does Inkling 2 Small cost per coding job?

A small fix of 500,000 input tokens and 100,000 output tokens costs about $0.25 in API fees. A standard feature of 2 million input and 500,000 output tokens costs about $1.10. These are illustrations on the published Tinker rates.

Do I need to pay for Tinker?

Tinker serverless inference is pay per use, with no subscription published. Production use is on a waitlist, and the serverless tier is in beta.

Can I self-host Inkling 2 Small?

Not yet. The weights are closed today, with an Apache 2.0 release promised for next week after safety testing. Until then, Tinker is the only way to reach the model.

What context window do I get on Tinker?

Tinker serves 64K and 256K variants today. The model card lists up to 1 million tokens, but that full window is not available by default on the platform yet.

Is $0.30 per million tokens really that cheap?

Yes, against the frontier models. The same small fix costs $4.00 on Claude Opus 5.5 and $2.00 on GPT-6.1 Sol in API fees. But your time, not the API fee, is the real cost of the service.

What if the model writes bad code?

You review every line, run the tests, and fix it before delivery. Price in one revision round, and charge for more. The model is a draft writer, not a delivery service.