I also use Claude Max with my company account, but that's strictly a company account after all.
I do occasionally test it for personal use, but I can't use it for personal projects (and shouldn't).
Looking at my usage pattern so far, even 2x MAX uses up my quota too quickly, so it's been quite burdensome to use with my own money.
Then I came across this video.
It's a recently trending Kimi-K2.6 test video.
It's outperforming the Opus 4.7 model in almost every benchmark.
Actually, it seems like almost all models are quite similar now, each just having different characteristics.
This one also looks quite attractive.
In the video, they run it on the cloud through HuggingFace, and compared to OpenAI, the price is almost 1/10th. The performance is also very solid.
You can run it locally too, but since it's 1.1T, it would be difficult for most individuals to run it (if you have the capability, it might be worth trying ㅎㅎ).
I'm thinking of trying it as a secondary option.
I could run simple tasks locally with something like Qwen3.6 35b, use that for tasks requiring larger context occasionally, and then maybe I wouldn't even need Claude.
