The cloud provider on Wednesday released an experimental version of its Gemini 2.0 Pro AI model and updated its reasoning models.
The experimental version of Gemini 2.0 Pro is best optimized for coding performance and complex prompts. It has a better understanding of and ability to reason about world knowledge than the original AI model, the vendor said. It is available in Google AI Studio, Vertex AI and the Gemini app for Gemini Advanced users.
Google also released Gemini 2.0 Flash-Lite, a new low-cost model, in public preview in Google AI Studio and Vertex AI. The model has a million-token context window and multimodal input capabilities. The vendor said it can generate one-line captions for around 40,000 different photos for less than a dollar in Google AI Studio's paid tier.
Flash-Lite is a version of Google Flash 1.5, which the vendor introduced in 2024.
The cloud vendor also made Gemini 2.0 Flash Thinking Experimental, its updated reasoning model, now generally available using the Gemini API in Google AI Studio and Vertex AI.
Google said all the models will feature multimodal input with text output, and more multimodal outputs are expected in the coming months.
The new releases follow many new products from competitors such as OpenAI, which also updated its reasoning model with o3-mini late last month. They also come as the AI model war heats up after the instant success of Chinese AI startup DeepSeek's R-1 models, also released in late January.
Amid the feverish reception of DeepSeek's free, open source model, Google made Gemini 2.0 Flash available in the Gemini app on Jan. 30, just over a week after the DeepSeek release.
Google is apparently also thinking about geopolitical tensions and the AI arms race with China. On Tuesday, it updated its AI Principles by lifting its own self-imposed ban on pursuing the use of AI and other technologies for weapons or that can cause harm. Google is also dropping diversity targets in hiring, the Wall Street Journal reported Wednesday, another sign that the tech giant is aligning with some of the ideological positions of the administration of President Donald Trump.
Google's leadership position
The change in its AI principles raises questions about Google's position in the AI race and whether it plans to take a more aggressive approach to developing advanced generative AI technologies, like one of its chief rivals, OpenAI.
"They've certainly been more aggressive in the last year or so," said Gartner analyst Arun Chandrasekaran. He added that Google's success with generative AI was evident in the past year since the cloud provider switched in 2024 from the former PaLM family of models to the Gemini line. The vendor also met success with its release in December of NotebookLM Plus, a generative AI summarization tool aimed at enterprises.
"They've certainly shown a level of intent and aggression in the market in the last 12 months that we didn't see as much in 2023 or even early 2024," Chandrasekaran continued.
Model specialization and price
With the Gemini 2.0 Pro and Flash line of models, including Flash-Lite, Google highlighted its strategy of adapting reasoning models for specific purposes. For example, its emphasis on coding shows that Google's target is to help developers build more production-scale applications, Chandrasekaran said.
"There's also some level of specialization ... with these models that were not very apparent when the reasoning models came out a few months ago," Chandrasekaran added.
Google also emphasized price. The low-cost Flash-Lite model costs $0.075 for every million token inputs of text, image, video and audio, and $0.30 for every million text output tokens. The cloud vendor offers a free tier for Flash-Lite.
Cheaper and better has to be there for this all to work and for generative AI to take off.
Mark BeccueAnalyst, Enterprise Strategy Group
By comparison, OpenAI's o3-mini (its lowest-cost model) costs $0.55 per one million cached input tokens and $4.40 per one million output tokens.
"This is a step in the right direction," said Mark Beccue, an analyst at Enterprise Strategy Group, now part of Omdia., which is owned by Informa TechTarget. "Cheaper and better has to be there for this all to work and for generative AI to take off. We've got to take that price down."
The emphasis on price underlines a problem enterprises face with adoption. Often, applications of generative AI models do not justify the cost. Therefore, by making models cheaper, Google and other generative AI vendors are encouraging more enterprises to adopt the technology.
"The fact that they're trying to price this thing, meaning the [advanced] frontier models, in a way that makes them more appealing for companies not just to use these for basic proof of concept and experimentation but also to use them in production is where the market is going," said Bradley Shimmin, an analyst at Omdia. "Google is definitely following the market trend."
For Google, the challenge will be to show how all its models differentiate themselves and also work together, Chandrasekaran said.
"How do you sell a premium product when you have a lower-priced product?" he said. "If they want all of these models within the model families to do well, they have to have clear guidance to customers and a clear selection process in terms of how customers are selecting the right model for the right use."
Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence
