Hy4 preview

“In backend Agentic Coding tests, previously Hy3 only scored 16 points, but this time Hy4 actually scored 10,779 points!”

“After running HY4 for a few minutes, the result was that the code was written wrong.”

“HY4 is completely unusable. I don't know how you all use it, but when I run a task with a few hundred lines of code, it freezes every time, and it's extremely slow.”

“Claims to be free, but always busy.”

“Because they just hastily trained it and served it up [crying with laughter].”

“An improvement over HY3.”

“Hy 4 Preview is cracked.”

“Currently HY4 feels like it's catching up.”

“The results can only be described as average. It's fine for competing with welm, but competing outside would be quite difficult.”

“You ask a task and there are thousands of people ahead, then the queue fails.”

“Can completely dominate Deepseek and Qwen's Flash models. Saying it surpasses K3 is a bit of an exaggeration, but its capabilities are genuinely top-tier.”

“This software doesn't have cache hit discounts.”

“I'm very satisfied with the intelligence level for solving complex and large-scale problems.”

“This model is pure garbage. I asked it to write a futures auto-trading script, and it straight up wouldn't run. After 100 rounds of Q&A, it was completely unusable in the end.”

“Tried HY4 for a bit, didn't feel it was particularly powerful. After all, it's been a short time, still need to spend more time using it. But it is indeed faster and more detailed in analysis compared to HY3.”

“Studied the exam questions too hard, once they throw a new question it falls apart [doge]”