There was once a time when you'd only hear "technical people" in the traditional sense talk about coding or building stuff. No-code building tools always lingered around, but compared to what we have now, the tools that existed back then were simply toys. Now, even the most non-technical person you might have around is constantly talking about the app they're "vibe coding" on the weekend, the SaaS they're spinning up with Cursor, or the side project they finally shipped because an LLM wrote 90% of it for them.

AI-assisted coding tools have effectively removed the barrier that once existed when it came to coding, and Claude Code is the tool that, for me (like many), sits at the top of the pile. I've cycled through all the alternatives, and I always find myself coming back to Claude Code. The way it reasons through a codebase, plans multi-step changes, and actually ships working code with minimal hand-holding is something else. Anthropic has the most impressive LLMs when it comes to coding, and Claude Code is the harness that squeezes the most out of them. And yet, despite all of that, I'm not entirely sure I can recommend it.

The limits are brutal, and they keep getting worse

And I'm on the $100 plan

I wrote an entire article about this just a few days ago, but the biggest reason why I'm struggling to recommend Claude Code to anyone nowadays is the brutal limits it comes with. Plus, Anthropic seems adamant on making the paid subscriptions worse, and while I can't blame them entirely for it because demand has clearly outpaced what their infrastructure can comfortably handle, that doesn't make the experience any less frustrating for the people actually paying for it.

I'm subscribed to Claude's $100 Max tier, and I've tried using the $20 Pro tier. I hit limits on the former, and it's the plan I've been daily-driving for months, so I can't imagine how frustrating it must be for someone on the Pro tier trying to get any real work done. If I'm running out of usage on a plan that costs five times more, the people paying $20 a month are basically getting a glorified demo (well, people say that a $20 plan is now the new "free" tier). Unfortunately, it's not like the Pro tier was always this restrictive. The limits have tightened over time, and I feel like they keep getting worse. This is despite the fact that Anthropic keeps doubling rate limits and also recently announced a partnership with SpaceX to secure more compute.

Just yesterday, I was working on improving a project for a course I'm taking this semester, and I ended up hitting my hourly limit just an hour and a half into my session. I wasn't really doing anything particularly intensive — just some back-and-forth refactoring. I was then locked out for the next five hours, with no way to continue unless I shelled out for extra usage on top of what I'm already paying. I ended up jumping to Codex, and despite being on the $20 plan over there, I managed to get a lot more done!

This has been happening more and more frequently over the past few months, to the point where I've started planning my coding sessions around when my limits reset. That's a ridiculous way to use a tool that's supposed to make you faster, not force you to schedule your workflow around an arbitrary timer.

Claude Code alternatives keep getting better

You have options now

Given that Claude's limits have been frustrating me a fair bit lately, I've been exploring alternatives more and more. Now, for a very long time, Claude Code was truly the best terminal-based AI coding tool out there, with nothing else even coming close. We had a lot of coding agents with IDEs, and a few full-fledged agentic IDEs showing up too, but nothing that quite matched the experience of a proper terminal-based agent. However, that's no longer the case, and the space has gotten significantly more crowded over the past few months.

OpenAI's Codex is an excellent example. When I first tried it out when it was fairly new, I was disappointed and switched back to Claude Code immediately. However, I've been leaning toward it more and more lately, and I keep walking away wondering why I haven't been using it more often. OpenAI has been launching banger after banger feature, and Codex has quietly become a really compelling alternative. The recent additions, the model improvements, and the fact that it's actually included on the free plan too all add up to something that's genuinely hard to ignore. Codex's limits are also nowhere near as bad as Claude Code's are right now.

Then there's OpenCode, which is the alternative I'd point most people toward right now. It's completely free, completely open-source, and the biggest selling point is that it lets you bring your own model. You can hook it up to most popular LLMs out there or even a local LLM running on your own machine. OpenCode has almost every feature that makes Claude Code well, Claude Code, including a Plan mode, Skills, agents, MCP servers, and so on.

The harness might not be as polished as Claude Code's, but you lose some to gain some. What you get in return is a level of flexibility that Claude Code simply can't offer at the moment. You're not tied to a single provider, you're not at the mercy of whatever pricing changes Anthropic decides to roll out next, and you're not going to find yourself locked out for five hours in the middle of an important task. If a new model drops next week that's better at coding than Claude, you just point OpenCode at it and keep going!

There are other alternatives worth mentioning too, and my colleagues here at XDA have constantly talked about them. That said, Codex and OpenCode are the two I keep coming back to when I think about what I'd genuinely recommend over Claude Code today.

There's also the closed-source problem

Open-source isn't just a buzzword

Now, this isn't an issue that really affects me specifically, since I don't work as a software developer professionally. However, with how great open-source Claude Code alternatives and local LLMs have gotten, it's become harder to ignore that Claude Code is, at the end of the day, a closed-source tool tied to a single company's cloud. If you're working on anything sensitive like proprietary code, client work under NDA, security research, or just anything you'd rather not send off to someone else's servers, Claude Code isn't really the best option.

Even if Anthropic's data handling is great (and I have no reason to believe it isn't), it's closed-source at the end of the day. This sort of brings me back to OpenCode, which is the best open-source Claude Code alternative I can recommend. It's fully open-source, which means anyone can read the code, audit it, and see exactly what it's doing. And because it lets you bring your own model, you can pair it with a local LLM running on your own machine and have a setup where your code literally never leaves your network.

Might be my favorite, but favorites can change

I'm a Claude fan through and through, but I've found it increasingly difficult to wholeheartedly recommend Claude Code to anyone over the past few months. The tool itself is genuinely excellent (arguably the best in its category), and Anthropic's models are still some of the most impressive out there when it comes to coding. Unfortunately, being the best tool in a category doesn't automatically make it the right tool to recommend, and we are where we are now.