I've tested free vs. paid AI coding tools - here's which one I'd actually use

akinbostanci / iStock / Getty Images Plus Follow ZDNET: Add us as a preferred source on Google. ZDNET's key takeaways Free AI chatbots handle small projects surprisingly well. Pro coding agents deliver serious productivity boosts for pro coders. Spend based on what you need to get the job done. The use of AI in coding has exploded in the past few years. What was an amazing curiosity in April 2023 is now impacting every programmer on the planet. As the IEEE (a professional organization for engineers) reports, headcount for young, early-career programmers has plummeted over the past two years. But more senior developers are utilizing AI to make themselves more productive. Then there are the AI tools themselves. They range from free AI tools that cost nothing, all the way up to nearly $1,000 per month for a range of professional AI tools. These tools use a considerable amount of computing resources, and their vendors charge accordingly. Also: The top 20 AI tools of 2025 - and the #1 thing to remember when you use them If you're a programmer or want to program, what does that mean? Do you need to spend the big bucks, or can you get away with a free solution? That's what I'll answer in this article. The short answer is, it depends on you. Let's dig in. Understand AI coding tools AI coding tools range from the very generally useful all the way to super in-the-weeds specialized tools for solving specialized problems. There are at least a dozen of these categories. Stay tuned to the end of this article, because I'll include a short glossary for each of those categories. But, for the purposes of this discussion, let's talk about three main categories: chatbots, AIs that live in the programming environment, and AI agents. Chatbots are like ChatGPT or Gemini. You type queries into a general-purpose chatbot, and it gives you back some code. This is the first way most of us became acquainted with AI programming. We dumped snippets of code into the chat prompt area, and the chatbot responded. Programmers use powerful text editors to write code. Usually, these editors are integrated into an IDE (integrated programming environment) that contains the editor, file management, source code control, debugging, sometimes a database access window, a terminal window, and more. Also: I'm an AI tools expert, and these are the 4 I pay for now (plus 2 I'm eyeing) If you know Photoshop, you know it has a bunch of different windows and tools, all designed to help you manage photos, all inside Photoshop. An IDE is like that, but for coding. AI tools are now becoming more integrated into IDEs. An open-source IDE from Microsoft called VS Code is the most common. Another one is Cursor, which is based on VS Code but has some native AI smarts. In each of these cases, the AI exists in various IDE tools and helps with code generation and command completion. A variation on this is the AI tools (like Claude Code) that work primarily through terminal windows. Since terminal windows are often found in IDEs, I'm including them in the same category. Then there are the AI coding agents. These are tools that can do a whole bunch of steps based on one single request or prompt. They can modify snippets of code, change whole sections of code, run terminal commands, upload and download changes to GitHub, and more. They are enormously powerful, but as Spider-Man's Uncle Ben says, "With great power comes great responsibility." We'll look at these more in a minute. But first, let's figure out what it is you need. What do you need? An analogy can really help answer this question. You want to move some stuff. What do you need? Well, that kind of depends on the stuff, doesn't it? What is the size of the project? If you're moving a big table saw from the local Home Depot to your workshop, you can probably ask a friend with a pickup truck to help you move it. The cost to you, worst case, is some gas money. On the other hand, if you need to move the contents of a house from New Jersey to Florida using a big rig with a 53-foot trailer, you're talking big bucks, because the cost of the team, the truck, road fees, and more adds up fast. So, when it comes to using AI for coding, are you doing a pickup truck's worth of coding, or do you need to run projects on the scale of sending big rigs across the country every day? As with the moving van example, there are cost economics driving it all, mostly based on resource constraints. After all, AI data centers are huge and cost billions of dollars. Someone's got to pay for all that gear, power, and land. Here are some of the ways various coding plans are resource constrained based on payment level: Session and token limits: Tokens are small units of text being processed. This is a measure of how many tokens can be processed in a request, as well as how long the request can run. Tokens are small units of text being processed. This is a measure of how many tokens can be processed in a request, as well as how long the request can run. Rate and usage limits: These are limits based on how many requests per minute, hour, or day, how much compute you can use, and how many concurrent sessions you can run at a time. These are limits based on how many requests per minute, hour, or day, how much compute you can use, and how many concurrent sessions you can run at a time. Model access and capability differences: As we've come to know, different AI models perform differently. AI vendors may limit lower-tier users to models that are less performant, require fewer resources, or have already been scaled. They may also gate the features you can access, like the ability to work across multiple files. As we've come to know, different AI models perform differently. AI vendors may limit lower-tier users to models that are less performant, require fewer resources, or have already been scaled. They may also gate the features you can access, like the ability to work across multiple files. Integration and environment limits: Lower-tier plans may limit integrations, like working only in VS Code, while more premium plans might support GitHub integration and other tools. They may also limit you to a single file in a code repository compared to accessing all the files in an entire repo. They also might not remember as much or have a way to retain context and project history. Lower-tier plans may limit integrations, like working only in VS Code, while more premium plans might support GitHub integration and other tools. They may also limit you to a single file in a code repository compared to accessing all the files in an entire repo. They also might not remember as much or have a way to retain context and project history. Latency, priority, and uptime: Lower-tier plans may run more slowly or be subjected to more waits. Sometimes free plans will experience various kinds of throttling or even downtime as overworked resources are assigned to higher-tier plans. My experiences I am not a full-time programmer. I do have programming chops, having been a computer science professor in a past life. So my use has varied based on what I'm trying to accomplish. Up until about two months ago, I was quite content to use in-chat ChatGPT coding support. I initially found that most of the ChatGPT coding worked fine in the free plan, although coding in the most obscure environments did baffle it. I've been paying for the $20-per-month plan because the paid plan offers more, and I'm constantly using ChatGPT for work projects in order to write about it. Also: Is ChatGPT Plus still worth $20 when the free version offers so much - including GPT-5? I do have a freemium security/privacy product that has a free open-source core with paid add-ons that mostly come close to covering my costs. I use it as part of my continuing practice to make sure I keep my programming chops up. Also: Google's Jules AI coding agent built a new feature I could actually ship - while I made coffee When Google's Jules and OpenAI's Codex came out, they were agent tools that worked solely in the GitHub repo. While both were fairly impressive, I didn't like the programming experience, so I didn't use them other than to write an article or two. (Disclosure: Ziff Davis, ZDNET's parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.) Also: I went hands-on with ChatGPT Codex and the vibe was not good - here's what happened But then Codex was updated to work inside VS Code. Even more interesting, OpenAI enabled Codex for those of us using the $20-per-month Plus plan. So I gave it a try. Also: I did 24 days of coding in 12 hours with a $20 AI tool - but there's one big pitfall The productivity wins were astounding. As the title says, I got the equivalent of more than three weeks of coding work done in 12 hours. But I ran into throttling that proved both annoying and kind of worrisome. Codex shut me down right in the middle of one of its runs, so it left my code in an unknown state. I didn't enjoy that. But I did enjoy the result. It not only helped me become amazingly productive, but it was fun. So I took a big step and spent $200 for a month of the Pro plan. Codex's project lead told me that I wouldn't experience throttling or shutdowns if I used the tool like a regular programmer (in other words, not spinning up hundreds of contexts at once). And I did. I did a four-day sprint, using Codex nearly 12 hours a day, and I wound up with four powerful new add-ons for my security product. Since my previous rate of add-on production was about one a year, I felt I could honestly say Codex gave me four years of productivity in four days. Also: I got 4 years of product development done in 4 days for $200, and I'm still stunned But then I turned the service back down to the $20-per-month Plus tier. That's because I can't sustain that kind of programming sprint on a regular basis. I rarely have that much free time, and, frankly, I've sufficiently advanced my product that I don't need to do more. Also: 10 ChatGPT Codex secrets I only learned after 60 hours of pair programming with it In fact, it took me another two or three weeks just to turn that code into products, record the videos, and create the online store pages. The product marketing took roughly five times longer than the coding. Frankly, if I didn't need the $20-per-month Plus tier for other aspects of my writing for ZDNET, I'd go back to the free plan. I just don't need the advanced coding tools that much, at least until I decide to do another big sprint. What did we learn? So what should all that tell you about your use of coding tools? The big takeaway is that it depends on what you're doing with them. When I switched from doing a little side programming to a major product development sprint, I paid more and got amazing results. But when I was done with that sprint, I downgraded my tools. If you're a full-time programmer, you (or your employer) are probably going to want to pay a few hundred dollars or more for a higher AI tier. The value-for-dollars ratio, considering how much coders cost and how much your time is worth, is well worth it. But if you're a hobby programmer, or you want a little help now and then, the free plan is mostly fine. If you want access to better models, you might want to pick one of the $20-per-month plans. But honestly, those free models were pretty bad in early 2023. In late 2025, they're really quite good. So you can get fairly far along without paying anything. Glossary of AI coding tool terms I track AI tools in order to write about them. Here's how I typically divide up the main categories: Editor IDE: Full development environments that are AI-native or deeply integrate AI to plan, write, refactor, and manage projects end-to-end. Full development environments that are AI-native or deeply integrate AI to plan, write, refactor, and manage projects end-to-end. Editor extension: Add-ins for IDEs/editors that bring AI chat, inline edits, test generation, or other assistive features into the editor. Add-ins for IDEs/editors that bring AI chat, inline edits, test generation, or other assistive features into the editor. Terminal agent: Command-line or shell-centric agents that read and write local files, run commands, and iterate from the terminal. Command-line or shell-centric agents that read and write local files, run commands, and iterate from the terminal. Cloud coding platform: Browser or cloud environments that provide AI-assisted coding, building, and deploying without local setup. Browser or cloud environments that provide AI-assisted coding, building, and deploying without local setup. GitHub automation: Bots or apps that act within GitHub to automate tasks like triage, labeling, PR prep, and policy enforcement. Bots or apps that act within GitHub to automate tasks like triage, labeling, PR prep, and policy enforcement. PR review: AI that reviews pull requests, writes summaries, suggests diffs, enforces guidelines, or proposes follow-up commits. AI that reviews pull requests, writes summaries, suggests diffs, enforces guidelines, or proposes follow-up commits. Code completion: Tools focused on inline suggestions and autocompletion as you type, often with chat and quick edits. Tools focused on inline suggestions and autocompletion as you type, often with chat and quick edits. Security/quality fixer: AI that detects vulnerabilities or code failures and proposes targeted, reviewable fixes. AI that detects vulnerabilities or code failures and proposes targeted, reviewable fixes. Documentation assistant: AI that generates or maintains READMEs, inline docs, API references, and explainer content from code. AI that generates or maintains READMEs, inline docs, API references, and explainer content from code. Snippet/knowledge manager: AI that captures, organizes, retrieves, and explains code snippets or developer knowledge. AI that captures, organizes, retrieves, and explains code snippets or developer knowledge. Self-hosted/on-premises server: Solutions designed to run on your infrastructure for privacy, control, and compliance. Solutions designed to run on your infrastructure for privacy, control, and compliance. Repo-aware assistant: Assistants that index your repository to provide context-aware chat, multi-file edits, and code navigation. Assistants that index your repository to provide context-aware chat, multi-file edits, and code navigation. Other: Useful AI developer tools that don't fit above or legitimately span multiple categories without a clear primary. There you go. You don't need to know all these terms, but if you do, you probably also know how much AI you need for coding. If you don't know most of these terms, you can probably get by with a free AI plan and use it to increase your programming chops or just help you write some code. Just remember: AI can be incredibly stupid. Regardless of how much you spend, it's up to you to guide it. It will help, but it's far from perfect. Even if you're new to this coding thing, don't assume what the AI recommends is more right than you are. If it seems like the AI is going over the top, down a rabbit hole, or has lost its mind, it probably has. Listen to your gut. Have you tried using AI tools to help you code? Do you stick with free chatbot options like ChatGPT or Gemini, or have you invested in a paid coding agent or IDE-integrated tool? What trade-offs have you noticed between cost, capability, and reliability? And if you've tried both free and premium tiers, where do you think the best balance lies? Let us know in the comments below. You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV. Get the morning's top stories in your inbox each day with our Tech Today newsletter. Featured