--- title: "ChatGPT vs Claude" description: "One person's experience of running both for a while. I get good results out of both, so the things that tip me either way are smaller and a bit arbitrary." date: 2026-09-11 tag: ai author: Lloyd Rowat canonical: https://grayforgelabs.com/posts/chatgpt-vs-claude --- # ChatGPT vs Claude *2026-09-11 · ai · by Lloyd Rowat · [Grayforge Labs](https://grayforgelabs.com/)* This is all just how it's gone for me, so take it as that. I've been using ChatGPT and Claude side by side for a while now and I get good results out of both. Research, debugging, drafting, talking through something I haven't worked out yet: whichever one I take it to, I usually come away with what I needed. It hasn't always felt that way. When I wrote about [a weekend spent inside Anthropic's tools](https://grayforgelabs.com/posts/claude-ecosystem-fortress) back in April, Claude felt clearly out in front. I don't feel that now. With GPT-5.6 Sol over the summer and then GPT-6 Astra at the start of this month, whatever gap I thought was there has closed, and set against Opus 5 and Fable 5.1 they feel comparable to me now, and both work. So the things that push me toward one or the other are smaller than that, and a bit arbitrary. ## Claude Talks More Claude is generally wordier. I'll ask something simple and get back headings, a summary, and a section I didn't ask for, and I'm scrolling around looking for the part I wanted. When the answer is going to someone else that extra structure is useful. When I just wanted to know one thing it's a bit much. ## Speed ChatGPT is fast. Scary fast sometimes, to the point where I'm briefly suspicious of the answer, though it's usually fine. I've no idea how much of that is the model and how much is everything around it, and I haven't tried to work it out. What I do notice is that it changes how I ask. When a reply takes a while I plan the question, cram everything into one prompt, try to get it right first time. When it comes back instantly I stop doing any of that and just talk to it. Claude isn't slow. It's slow enough that I behave differently, which is not quite the same complaint. ## Limits Claude stops me more often. It runs on five-hour windows and I'll hit the end of one a couple of times a week, usually right when things are going well. I don't hit that wall on ChatGPT, or at least I haven't yet. I can't really back this up. I use the two differently, most of the heavy lifting goes to Claude, and I've no good way of comparing them fairly. It just feels like more headroom over there. One thing that's probably doing some of the work is that ChatGPT keeps Codex separate from normal chat, so a long stretch of coding doesn't stop me asking something unrelated later on. With Claude it all comes out of the same pot. ## The Apps ChatGPT does images and Claude doesn't, and that's mattered more than I expected it to. Nothing serious, it's just there when I want it. The desktop apps I'd call a draw. Both are a chat window that syncs with my phone, and I never think about which one I'm in. ## Coding On the code itself I've no preference. Both give me something I'm happy to work with, and I don't catch myself wishing I'd asked the other one. What I do prefer is the Claude app for actually doing the work, and on my phone especially. Starting a task from mobile, having it run in a cloud environment I never had to set up, and coming back later to a pull request is the bit I'd miss if I stopped. Codex is the weaker half for me there, mostly on connection. Remote sessions drop often enough that I expect it now, where Claude's tend to stay up. Reconnecting is easy enough, it's working out where it had got to before it went quiet that irritates me. Some of that is probably just habit, and the same pull I noticed back in April. If I had to drop one I'd probably keep ChatGPT, mostly on speed and on not being told to wait. It's not a strong opinion and I'd understand anyone landing the other way. The Claude app is the part I'd actually miss, which is enough to keep me on both. ## Update: Opus 5.5 *Added September 23.* This post is twelve days old, which is apparently long enough in this industry for it to need an update. I've changed my mind. Opus 5.5 came out this week, and I've spent the past day using it and Sol-6 side by side. For what I do, I've preferred Opus 5.5. It's much nicer to use, and it fixes some of what bothered me about Opus 5. This isn't a knock on Sol-6, which is still very capable, and the two labs pushing each other this hard has been good for anyone using either one. It's the same kind of small, personal call as everything else in this post, only this time it lands the other way. A day is not long, so I'll see whether it holds, but if I had to drop one today I'd keep Claude.