\n<\/p>\n
Sidhant Bendre canceled his AI startup’s ChatGPT subscription and switched to Claude last year.<\/p>\n<\/div>\n<\/li>\n
He said Claude performs better and is among a wave of users switching over to Claude from ChatGPT.<\/p>\n<\/div>\n<\/li>\n
ChatGPT’s parent company, OpenAI, has missed revenue and user growth goals, according to reports.<\/p>\n<\/div>\n<\/li>\n<\/ul>\n
This as-told-to essay is based on a conversation with Sidhant Bendre, a 26-year-old cofounder of Oleve, an AI-driven consumer software portfolio company, based in New York. His words have been edited for length and clarity.<\/em><\/p>\n I used ChatGPT for roughly two years before moving over to Claude as my preferred model and ending my company’s ChatGPT subscription.<\/p>\n When we launched our company, Oleve, we were already using AI, so we’ve always relied heavily on it in our workflow, from coding to marketing to our hiring process. ChatGPT was our go-to model. After Anthropic released its 4.5 model suite for Claude last fall, that changed.<\/p>\n We weren’t pushed away from ChatGPT; we were pulled into Claude.<\/p>\n We didn’t have any major frustrations with ChatGPT that forced us to switch, but I think one of the promises of AI is to speed up our work, and Claude let us move faster because we spent less time correcting it.<\/p>\n When we were primarily using ChatGPT, I noticed that the content it generated didn’t sound very human. It would overuse emojis where I wouldn’t, and it felt really forced. I haven’t used ChatGPT rigorously for a while now, but whenever I go back to test prompts with the model, I still get those overly verbose responses.<\/p>\n Claude is better at mimicking human writing. Even before Claude’s 4.5 suite came out, we’d heard students talk about how the model was so good at mimicking their writing style after they fed it examples.<\/p>\n With Claude’s coding, we were able to automate a lot of our development time with blueprints we already had. Now we can focus more on the product itself rather than spending so much energy on build time.<\/p>\n The biggest difference between the models seems to be parsing context and considering nuance.<\/p>\n If I feed Claude a large research document and I’m looking for something small, it almost always understands when to be concise, while providing the appropriate context and highlighting what I’m asking for more details on.<\/p>\n Sometimes it felt like ChatGPT was overcompensating for poor judgment about what to focus on by spitting out more content. If I’m using it at all now, I usually have to ask it multiple times to summarize or reinstruct it to be more concise.<\/p>\n We’ve integrated AI into more of our workflows after switching to Claude. It made more sense to invest the time and energy it takes to do that when I know the payoff is worth it.<\/p>\n Recently, Claude has been giving us its own set of issues with bugginess. Some of our messages or chats have been disappearing, and this has been frustrating.<\/p>\n It’s not perfect. I feel like there are still hallucinations I see in the model’s output every now and then, and that makes my trust in Claude crumble a little bit. I have to push back in these scenarios to self-correct, even with web search on their new Opus 4.6 model.<\/p>\n I haven’t really been paying attention to the larger OpenAI vs Anthropic debate. If ChatGPT launches a product suite that provides significant value to me, I would try it. I don’t feel attached to Claude to the point where I wouldn’t try something else.<\/p>\n I think that the biggest thing for me has been that Claude feels so much more like what the promise of AI is supposed to be. All the models have provided significant value, but it feels a lot more like it’s taking away the busywork from my day-to-day and giving me back time to think big picture.<\/p>\nRight off the bat, Claude had fewer bugs generating code than ChatGPT<\/h2>\n
The way the models understand nuance is the biggest difference<\/h2>\n
Claude has given us its own set of problems<\/h2>\n
I’m still a big fan of Claude’s suite, but if something better comes along, I’d switch<\/h2>\n