Since my local Qwen 3.8 did most of the work the recent weeks I thought about canceling my "mistral-medium" Pro subscription. But when I started the vibe-cli and saw that the context size increased to 800k, noticed that they serve GLM-5.2 now, so I'll probably keep the sub. The pricing (for now) gives me a good large quota included in my Pro subscription and the model seems very fast. Via API (hits a smaller, but different quota bucket) it even gives 1M ctx. Built a small proxy to wire it into Android Studio to work around Googles sub-par third party support.
I mean of course it is a Chinese model (like Qwen), but a quick censorship test tells me it is at least not extreme. Just like Qwen, it happily critizes the Tiananmen Square massacre and Tank Man aswell as makes fun of the Winnie Pooh ban in the PRC.
1 comment
[ 0.23 ms ] story [ 14.1 ms ] threadI mean of course it is a Chinese model (like Qwen), but a quick censorship test tells me it is at least not extreme. Just like Qwen, it happily critizes the Tiananmen Square massacre and Tank Man aswell as makes fun of the Winnie Pooh ban in the PRC.