Free AI for programming: what is actually available

The search "free AI for programming" usually means one of two things: someone wants to try how it works, or someone wants to work every day and not pay. Free options fit the first goal. For the second they almost always hit limits. Below we go through the free paths, what they cost you in other ways, and what a paid minimum costs if free is not enough.

Free AI for programming: which options exist

There are four, each with its own price.

Trial limits at services. Many services give a limited number of requests without payment. The terms change often and we do not know them: check the current limits on the service's own site. This is enough to get the idea, but usually not enough to work on a project: the limit ends in the middle of a task.

Local models. Open models can be run on your own computer with tools like Ollama or LM Studio. You pay nothing per token, but there are other costs: you need a GPU or a lot of RAM, the model usually fits only in a reduced form, and speed depends on your hardware. We cannot tell you the quality in advance, so test on your own tasks.

Free extensions and plugins. Some editor extensions are free, but the extension itself contains no model: it connects to one, and that is the part you pay for or run into limits with. A free extension is not the same as free work.

Small bonuses. Some services give a small balance for an action. Zukko, for example, gives $0.50 for subscribing to the @zukkopro↗ channel: once per Telegram account, after linking Telegram in the dashboard. It is a trial balance to check the connection and look at spend, not a way to work for free. We have no regular free access.

What to do if you need the free option

If the goal is to learn and experiment, combine two paths: a small local model for simple questions and the online services' trial limits for harder ones. If the goal is steady work on a project, estimate the paid minimum. It turns out to be smaller than people expect.

AI for writing code: what a paid minimum costs

Billing is per token with no subscription: you pay only for tokens actually sent and received. Prices as of 18 October 2026, $ per 1M tokens (input / output):

The index is the Intelligence Index from Artificial Analysis. It reflects general model level, not the result on your task. Two models with similar indexes can handle your code differently, so run them on real requests.

One request to fix a function

Take a typical chat request: you paste a function and ask to fix it. Assume 3000 input tokens and 800 output tokens. Formula: input × input price + output × output price.

• deepseek-v4.1-flash: 3000 × 0.013 / 1M + 800 × 0.038 / 1M = 0.000039 + 0.0000304 = about $0.00007. That is roughly 14,400 such requests per $1.
• glm-5.3-flash: 3800 × 0.04 / 1M = $0.000152, about 6,600 requests per $1.
• claude-sonnet-5-5: 3000 × 0.4 / 1M + 800 × 2 / 1M = 0.0012 + 0.0016 = $0.0028, about 357 requests per $1.

Even the minimum $1 top-up gives more than ten thousand requests of this size on the cheapest model. For personal learning that lasts a long time.

An agent session

An agent that reads files and runs commands on its own sends the accumulated conversation to the model on every step, so spend grows faster. Assume 30 steps, an average context of 40 thousand tokens, so 1.2M input tokens and 30 thousand output, ignoring cache:

• deepseek-v4.1-flash: 1.2 × 0.013 + 0.03 × 0.038 = 0.0156 + 0.00114 = about $0.017.
• glm-5.3-flash: 1.2 × 0.04 + 0.03 × 0.04 = 0.048 + 0.0012 = about $0.049.
• claude-sonnet-5-5: 1.2 × 0.4 + 0.03 × 2 = 0.48 + 0.06 = $0.54.

For Claude in the Anthropic format (/v1/messages, Claude Code) the gateway sees the cache: reading costs 0.05 instead of 0.4 per 1M, and writing costs 0.5. The real figure depends on how much of the context comes from cache. More on the calculation in how much an agent session costs↗.

Supplier prices change. The current list is GET https://api.zukko.pro/v1/models or the Models page in the dashboard.

How to connect a cheap model to your editor

The idea: an extension or coding agent asks for an API address and a key. For OpenAI-format clients the base URL is https://api.zukko.pro/v1, the key looks like fetch_..., and the model name comes from the list above. A check from the terminal:

curl https://api.zukko.pro/v1/chat/completions \
  -H "Authorization: Bearer $ZUKKO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4.1-flash",
    "messages": [{"role": "user", "content": "Write a Python function that computes Fibonacci numbers"}]
  }'

Setup steps for different editors are in the beginner guide↗, and the current model list is in the models reference↗. To avoid spending more than planned, you can set a daily and monthly dollar limit on a key in the dashboard.

How to pay for the paid minimum

Top-ups start from $1. Options are SBP (the payment page adds an 8.5% fee on top, and the balance receives the dollar amount you requested), cryptocurrency and MasterCard cards. Topping up $1 and seeing how much a working evening actually uses is cheaper than choosing blindly.

FAQ

Is there a coding AI that is free with no limits?

We do not know of permanent unlimited free access to strong cloud models. Only local models are unlimited, but you pay with hardware and setup time.

What should a beginner pick?

Start with a cheap model like deepseek-v4.1-flash or glm-5.3-flash: for pennies you can ask thousands of questions. If the result is not good enough, switch to claude-sonnet-5-5 and compare on the same requests.

Is a local model better than a paid API?

You cannot say in advance: it depends on the model, hardware and tasks. Compare on several of your own requests, measuring both the result and response time.

Can I cap spending so the bill is not a surprise?

Yes. Each key has a daily and monthly dollar limit and a list of allowed models. Details are in the key limits reference↗.