Google has released what it calls a new “reasoning” AI model; However, it is still in the experimental stage and, according to our brief tests, there is definitely room for improvement.

The new model, called Gemini 2.0 Flash Thinking Experimental (a mouthful, of course), is available on AI Studio, Google's AI prototyping platform. A model card describes it as “best for multimodal comprehension, reasoning, and coding,” with the ability to “reason on the most complex problems” in fields such as programming, mathematics, and physics.

One to mail Logan Kilpatrick, AI Studio's product lead, called the Gemini 2.0 Flash Thinking Experimental in X "the first step in (Google's) reasoning journey." Jeff Dean, chief scientist of Google DeepMind, Google's artificial intelligence research division, said: in question He noted in his own post that the Gemini 2.0 Flash Thinking Experimental is "trained to use thoughts to strengthen its reasoning."

“We see promising results when we increase the inference time calculation,” Dean said, referring to the amount of computing used to “run” the model when evaluating a question.

Built on Google's recently announced feature Gemini 2.0 Flash The design of the model, Gemini 2.0 Flash Thinking Experimental appears to be similar to OpenAI. o1 and other so-called models of reasoning. Unlike most artificial intelligence, reasoning models effectively control themselves, which helps them avoid some of the pitfalls that normally trip models up.

As a disadvantage, reasoning models often take longer to arrive at solutions (usually seconds to minutes).

When given a stimulus, Gemini 2.0 Flash Thinking Experimental pauses for a few seconds before responding, considering a series of relevant cues and "explaining" its thinking along the way. After some time, the model summarizes what seems to be the best answer.

This is what should happen. When I asked Gemini 2.0 Flash Thinking Experimental how many R's there were in the word "strawberry" he said "two".

Your mileage may vary.

inside Following the launch of o1something happened explosion The number of reasoning models from rival AI labs, not just Google. In early November, DeepSeek, an AI research firm funded by quant traders, launched a preview of its first reasoning model. DeepSeek-R1. Same month, Alibaba's Qwen team explained He claimed to be the first "clear" rival to o1.

What opened the floodgates? The first is the search for new approaches to improve generative AI. As my colleague Max Zeff did recently reported“Brute force” techniques for scaling up models no longer provide the improvements they once did.

Not everyone is convinced that reasoning models are the best way forward. They tend to be expensive thanks to the large amount of computing power required to run them. While they've performed well in benchmarks so far, it remains to be seen whether their reasoning model can maintain its current pace of progress.