GPT-6 Luna was the only model to fail a network scanner dashboard task, reviewer says
The author tested several models on a network scanner dashboard task, found GPT-6 Luna MAX the only complete failure, and suspects GPT-6 Sol is a regression from 5.6 Sol.
I've been a Luna hater since the launch of GPT-5.6 Luna, so my experience with GPT-6 Luna was brief. I only tried it once, giving it the same task I gave Opus 5.5, Astra 6, MiMo V2.6 Flash, and Muse Spark 1.3 - building a network scanner dashboard. GPT-6 Luna (MAX) was the only model that completely failed the task. For some reason, it built a Bluetooth-only scanner, while all the others handled Wi-Fi, Bluetooth, LAN, etc. Honestly, I think it's a pretty shitty model. There are plenty of better and more affordable alternatives. Using it for free users? I get it. Giving it to Pro users who are paying 5x more as a "daily driver" because the top-tier model, Astra, is too expensive? I really don't like that at all. Also, GPT-6 Sol itself might be a regression from GPT-5.6 Sol. With all the hype OpenAI built around the GPT-6 Sol release, I think they fucked up pretty badly, and DevDay might be their best chance to turn things around. But even then, Anthropic cooked with Opus 5.5, and I'm not sure what OpenAI could release that would really change the situation for me. A model that's better than Opus 5.5 with better limits? Sure. But I honestly doubt that's what we're going to get. Personally, if DevDay ends up being a bunch of stuff like "o" assistant, a new $500 subscription plan, hardware announcements, and similar stuff, that won't convince me to renew my current OpenAI subscription. I get it, geeks love hardware, but the vast majority of users probably won't care about it at all, myself included. I'd rather switch to Claude's $100 plan with its high limits and access to what I currently consider the best LLM available.