Jeff Dean, chief scientist at Google DeepMind, said that the model has been given additional computational power, saying that for X, “we see that increasing the computation in inference time yields promising results! ” he wrote. This model works by pausing and considering multiple relevant prompts before providing the answer that it deems to be the most accurate.
Since OpenAI entered the “inference” space with o1-preview and o1-mini in September, several companies have rushed to achieve feature parity with their models. For example, DeepSeek released DeepSeek-R1 in early November, while Alibaba’s Qwen team released its own “inference” model QwQ earlier this month.
Although some people claim that inferential models can help solve complex mathematical or academic problems, these models are not suitable for everyone. Although it has shown good performance in some benchmarks, questions remain about its real-world usefulness and accuracy. The high computing costs required to run inference models have also led to rumors about their long-term viability. For example, OpenAI’s ChatGPT Pro costs $200 per month because of its high cost.
Still, Google seems to be seriously pursuing this particular AI technology. Logan Kilpatrick, an employee at Google’s AI studio, called it “the first step in the inference journey” in a post about X.
