https://neurokov.com

So I made an agency named Neurokov https://neurokov.com and basically its goal is to automate buisness which inlcudes marketing, workflows and basically anything related to automation. So should I stick with this is specialise in one specific automation like marketing?

Sonnet 5.5, this AI model we saw a few days ago, was the second-best model. I think it has a score of around 56 on the Artificial Analysis Intelligence Index, and it costs $2 input and $10 output. All of us, including me, were wondering: why the hell would I use GPT-6.1? It is the same price, but it is worse. The thing is, cost per task matters significantly too. I mean, that model is better than Astra. But that model is also way more costly. It is not actually cheaper just because it is $2 input and $10 output. That is not the whole story. You should keep that in mind. I just saw that there was a report and some insights showing that Sonnet 5.5 uses around 60% more reasoning tokens at the highest level measured. So my analysis is that I think these guys are basically not upgrading the model internally that much. It is not some free lunch. I think they are just making it reason really, really much more. They just upgraded the model and, for max effort, made it reason a lot. So it is just burning through tokens. It is burning really, really many tokens. That is the thing. Their advertised price is not the whole story. Cost per intelligence task is a way bigger story. GPT-6.1 Max costs around $0.70 per task, while Sonnet 5.5 Max costs around $7. That is just huge. I think that is around a 10x difference in price for just around a 6-point increase. And another side point: for less than a dollar per task, you can get Opus 5.5, which is better than Sonnet 5.5. So at that point, it is just not really worth using Sonnet 5.5. But that is more of a side point. Remember, the pricing per token is basically the same, but the cost per task is way more. So input and output token prices are not the whole story. What matters is the cost per task and how many tokens the model actually uses. Because many times we are going to have models that cost one cent input or 10 cents output, but if they are using so many tokens, then that advertised price just does not tell you the full story.