The recently released 3.5 is much smarter than the Qwen 3 235b, which is also less efficient. Why it hasn't been deprecreated yet is beyond me. Could you pls add Qwen 3.5 thinking/non-thinking?
The qwen models have historically benchmaxed and been very quirky. They end up being less useful than other models in the same class, and they seem to be less innovative than the other Chinese labs like Deepseek, Moonshot, Minmimax, or Z-AI. I don’t think that’s an excuse though. The qwen models run fast, and they are one of the only labs still making models small enough to run on your own hardware. There’s honestly no good reason to be this late in adding support.
Indigo Peacock
Yeah I’d really like this one especially as this is one of the only open models that performs pretty high on bullshit bench. I know it’s not the MOST important benchmark but I would like the option of a low sycophancy model other than Claude since they’re pricey and it’s the next best
Apricot Astronaut
yea, normally they’re quick to add models - felt like they were automatically prioritizing addition of new models. (0-1 days after release if western model.)
but this is the first latest model they delayed for quite some time.
Log in to comment and vote
Comments4
Jade Custard
The qwen models have historically benchmaxed and been very quirky. They end up being less useful than other models in the same class, and they seem to be less innovative than the other Chinese labs like Deepseek, Moonshot, Minmimax, or Z-AI. I don’t think that’s an excuse though. The qwen models run fast, and they are one of the only labs still making models small enough to run on your own hardware. There’s honestly no good reason to be this late in adding support.
Indigo Peacock
Yeah I’d really like this one especially as this is one of the only open models that performs pretty high on bullshit bench. I know it’s not the MOST important benchmark but I would like the option of a low sycophancy model other than Claude since they’re pricey and it’s the next best
Apricot Astronaut
yea, normally they’re quick to add models - felt like they were automatically prioritizing addition of new models. (0-1 days after release if western model.)
but this is the first latest model they delayed for quite some time.
Mark Florkowski