Why Open Weight Models and Ollama Beat ChatGPT and Claude

Leaning Into Open Source Models

With the variety of models being released almost weekly at this point, and the fact that almost everybody is doing about the same thing, I would recommend leaning into open source models. That's the thing. We've kind of been brainwashed, or I wouldn't say brainwashed, but I think the normies or most people who aren't really interested in doing the dirty work know which model is the best. These individuals are the ones who are going to fall into the ChatGPT, Claude, and Gemini traps. There's nothing wrong with that since they do make good models, but I would have to say we need to do a better job at making the open weight models more accessible.

One thing I think we can do about that is use and hype up tools like Ollama. You know, I didn't really understand the importance of using open weight models. Well, I guess I did. I'll take that back. I did, considering I really enjoy being able to run an LLM on my device and not worry about the data going anywhere. On top of that, being able to use the cloud models when I want to and get that extra buff when it comes to the intelligence at a decent price is great.

My Experience: Anthropic vs. Ollama

I think I started using Ollama more this year when it came to OpenClaw. I got the $100 plan for Anthropic, and I got the $100 plan for Ollama. I think the Kimi K25 model came out right about the same time. I think we were still on 4.5 Sonnet, maybe just on 4.6. I said to myself, "You know what? We're gonna try out both of these and figure out which has the best results."

Of course, Claude and the Sonnet model were doing really well, but my usage was gone. I didn't feel like I got that much use out of my agent when I was using it with OpenClaw at the time. Then when I swapped over to Ollama, everything felt great. I didn't feel like I had to worry about my token usage. I definitely didn't have to worry about getting banned just 'cause that was a thing; they definitely changed the TOS, which was crazy. I just felt like I had a way better result. So, I just continued to use Ollama.

The Rise of Open Weight Models

Kimi K25 was good. GLM5 came out, and that was good. More open weight models came out, and they were better. Here I am, people are still talking about Anthropic banning people with OpenClaw, and they're like, "Ah, we're trying to find different models." I feel like I tried my hardest on X to say like, "Guys, you need to use Ollama." Now people are kinda seeing the truth and switching over to Ollama. I feel like I had known the truth since maybe mid-February. I had been using it since early OpenClaw release 'cause it was just pretty nice. I didn't feel like I had to worry about any token usage. I was getting like 30% max.

GLM 5.1

Now GLM 5.1's out, and this is great. This model is top tier, most definitely. I feel like it's able to get a lot of what the other models weren't able to do when it came to the open weight models. I liked Kimi K25, but I felt like when it came to programming, it just wasn't really there. When the MiniMax 2.7 model came out, it did a little bit better, but I found myself kind of going back to Kimi just 'cause Kimi was great. But I'll say with GLM 5.1 now, that right there is great.

Improving My Development

I was able to improve my development, most definitely. I have this project here, my blog, and it was able to help me with that with minimal error. Some things it'll mess up on, but for the most part, it's been able to do a fairly well job, just with some correction here and there. Sometimes it'll add a component when it doesn't need to. I always say it could be how I prompt it. If I'm not specific or I don't say, "Do something," it has to interpret it and then it might interpret it wrong, which will mess you up.

My favorite feature is that GLM helped me make the dictating tool that I'm using right now to make this blog post. It's just my chat. I just wanted to riff, and I'm gonna make a blog post about it 'cause this was just on my head.

The Moral of the Story: Use Ollama

The moral of the story is you need to use Ollama. When you start using Ollama, you need to use the models to make your own agent harness. Everybody's talking about agent harness. I'll probably make another blog post talking about that, and it'll just be me riffing and rambling.

If I wanted to wrap it up with a nice little bow, my experience from using both the Anthropic plan and Ollama is that I canceled my Anthropic plan, and I am sticking with Ollama. I think I'm gonna build most of my apps around Ollama from here on out just because they support open source. I like their product. I like their tools. I like everything about them. I like that I can use open source models. I like that people don't have to rely on someone's infra if they don't want to. Everyone and anyone can use my app based at their level and scale at whatever they need to. That's what I wanna do.

I like Ollama. Hopefully, you guys find it and use it too. You can use it for free. Some of their clouds will get a taste. Their $20 plan's pretty good. Their $100 plan is even better. I'm not sponsored, but if I could get affiliated, that'd be great. Your boy needs a new computer or something. I'll talk to y'all in the next post.