Cheap, Fast, or Smart? Evaluating Grok Code Fast 1

The pitch for the awkwardly-but-accurately named grok-code-fast-1 is: it's pretty good at code, it's cheap, and it's fast.

But how does it actually compare with the other cheap-and-fast models that you might want to consider? Let's take a look. (As always, you can explore the full results here.)

Speed

GCF1 is top of the class for raw tokens output. The trouble is that it's not smart enough to solve Power Ranking tasks without lots and lots of tries. So even though GCF1 can spit out almost twice as many tokens per second as Flash 2.5, Flash 2.5 actually finishes the tasks in less time.

This chart is actually the better one for GCF1 since it only counts solved tasks, meaning GCF1's time is summed across fewer tasks than are the higher-scoring models. Hit the link above to see what it looks like with All Tasks.

Cost

GCF1 looks a little better in the price/performance department, where Flash is significantly more expensive. (Although again, we're only looking at solved tasks here.) Still, GPT-5 nano is both smarter and cheaper.

Verdict

Cheap and fast is a useful niche to have competition in, so we hope xAI keeps trying! But for now, if you want the smartest fast model and you're willing to pay a bit more for it, you should use Flash 2.5. And if instead you're optimizing for the cheapest possible fast model, then you want GPT-5 nano.