The Thinking model has a higher hallucination rate than the Instruct model and tends to overlook details in long contexts.
#4
by
xiaoxiao218
- opened
This comment has been hidden (marked as Off-Topic)
不过很好的一点是，intruct模型在回答该问题的准确性上优于所有此前的开源模型，甚至是qwen3-max-preview和kimi-k2-0905
But the good thing is that the intruct model outperforms all previous open source models in answering this question, even qwen3-max-preview and kimi-k2-0905