Grok Code 1 Fast has been a smash hit since it’s launch, mainly because it was a free model with a peak of 240B tokens on one day:

This is the only thing I can think about when I read the new model’s information of Qwen3 Coder Next: that it’s a competitor to Grok’s coding model. Alibaba has dropped many great open weight models so far, but this feels a direct shot at smaller coding models. Especially when you look at the 3B active parameters.

It even matches the pricing of the Grok model for both providers:

The only thing that this model is missing so far is prompt caching at the providers that it’s hosted on, but I think that’ll be changed soon.

The model is fairly fast at catching information from a log for example in the Kilo CLI:

It also doesn’t do verbose outputs. The model is very direct. Where Grok Code 1 Fast usually burns through a lot of tokens just to answer something, this model just replied with this information with some highlighting of the issue where for the same prompt Grok seemed to come with a lot of other additional information like “The impact on the pipeline” because it failed…

The model is also fairly balanced on price to performance (benchmark vs size) according to their research post.

I do think the most interesting benchmark is the amount of agent turns it’s capable of:

Where other models seem to not be capable of that, which causes issues in longer running agentic tasks (for example, research and planning).

For the upcoming days, I’ll be trying out this model on more tasks to really understand what it’s capable of doing.