I just can't get over how good Qwen3 Coder actually is. It kind of makes me want to figure out how to wrap these open source models like Kimi K2 and Qwen3 Coder and future ones that come out with a fixed price monthly service that people can use with healthy limits. There will.

Preparing to go into August Eval runs, this one includes an additional test so scores will be higher. If there are any missing from this list you really want to see let me know and i'll add it. (Minus Kiro i'll add this one in the future when its more opened up, i'm too limited

Its been an amazing couple weeks. 1. Qwen 3 Coder was a massive surprise for me!.2. Claude Code sub agents opens up so many things I want to experiment with. 3. We got approved for up to 250k in Google Cloud credits including Gemini usage, which also opens up a ton of things. 4.

Anthropic keeps increasing rates but is down so much!!! Its getting a bit frustrating honestly.

Claude Code now has the ability to create custom sub agents. Use /agents and follow the steps. This unlocks so much, I’ve created a couple already and it’s working great. Big feature request is to be able to assign opus or sonnet to the sub agent.

I do know some people have posted negative things about this model, but I can only share my experience. But i'll say temperature matters a lot, as does what provider you are using, so test in your setup for speed, cost, and stability. I made a request to OR to hopefully improve.

Video link here:

Qwen 3 Coder is honestly a joy to work with. In RooCode with a stable provider in OR I get exceptional results. 1. It DOES have some of the same provider issues as Kimi K2, slow tps on some, as well as inconsistent results. 2. It DOES end up costing more than I anticipated due

I can't tell you how hyped I am about Qwen3 Coder, I spent the entire day solely coding with it, and it was actually enjoyable. Reminds me of the days when DeepSeek V3 first came out. It works amazingly in RooCode, you just have to set temp to 0.7. Video coming out in the.

Meet my pet Fluffly, this was built using the new Qwen3 Coder model in RooCode (where it works great). Should have a full review with evals out tomorrow morning!. It works in RooCode exceptionally well.It works in opencode fairly well. Debating if I should test out Qwen Code.

Here's a quick gif what I've been building. Behind the scenes is a very complicated agentic system:. 1. We use a combination of models 4o, Claude 3.7, Claude 4, 4o-mini, and Gemini 2.5 pro. 2. Machine Learning models that automatically train based on a particular stores data.

You'll never guess, but Claude is down again. It must be painful for the Anthropic team to deal with this constantly. At least its Frolicking

Seeing an enormous amount of people saying Claude 4 has degraded significantly. I don't think its been as terrible for me yet, but what I have noticed. What I've noticed very subjectively at the moment:.1. Shortcuts that didn't feel like they happened before. Example I can't fix

Video up now:

Video coming soon to discuss this in more detail, but my main points are:.1. Developers are not stupid let us choose when we need to. 2. Each model is almost like its own programming language where you learn to talk to it very differently. Curious if i'm in the minority here. .

I also understand the need to optimize cost, its clear to me the cost per messages are going up. In fact in Raleon i've seen our cost per user message go from $0.03 to $0.20 which is a substantial difference. The reason for this is Tool calling. We keep being told the cost of AI

Lets talk Cursor for a second, Auto mode is how you get the most requests for the $. The problem with this approach is that you can get wild variance. In fact across eval scores there is a swing of nearly 50%. Because it all depends on the models you happen to get. BUT I prefer

Model Selectors in AI Coding Assistants should always be an option. A few companies including AmpCode have doubled down on the idea that model selectors aren't needed. I appreciate bold stances, but I strongly disagree with this take. I am working on a full review + evals of

AI coding works better when you know the inner workings of what you are building… frameworks + language. I’m a noob at Rust for example and I find it painful to really guide AI effectively when working with Rust. While Node,C++,Java etc I feel like it gives me super powers. How.