Has anyone tested last night’s releases of Grok4.6 and DeepSeek V4 Pro? The official version of V4 Flash is already insanely powerful. How about the official V4 Pro? Which one is stronger compared to Grok4.6? I’m debating whether to buy the official DeepSeek API.
1 Like
v4 Pro probably messed up.
deepseek is a bit stronger, grok4.6 gets dumbed down really easily.
If you have a subscription, it’s okay—but even the subscription version gets dumbed down.
I’m currently using the company’s Heavy account. After using it for a day, it doesn’t seem much different from 4.5 
When it comes to the so-called “AI god models,” let’s just say—none of them are actually good!
DeepSeek-V4-Pro is just as much of a rip-off as Claude-Fable5. DeepSeek-V4-Flash offers more than 95% of the capabilities of the Pro version, but at a much lower price.
As for Grok, that’s not even worth mentioning—an American LLM contender that starts lashing out at users when it gets stumped. What do you think?
In terms of overall performance and cost-effectiveness, grok4.5 > ds v4 flash > grok 4.6 > ds v4 pro.
Grok seems pretty decent 
. I won’t comment on DS since I haven’t used it much, but subjectively, Grok feels a bit more enjoyable to use than GPT5.6 SOL. With 4.5, it felt slightly less intelligent, but its coding ability seemed pretty solid
So they’ve turned over a new leaf—the one who used to sit at the same table as Doubao.
Indeed, before 4.5, I treated it as a toy. When I heard that 4.5 was good at coding, I gave it a try. As it happened, my boss’s Claude account got banned, and I found Grok really enjoyable to use, so I enabled Heavy 
d, strong, right?