Jack Silva | Soup Images | lightrocket | Getty Images
Google On Wednesday, everyone released Gemini 2.0, its “most powerful” suite of AI models.
In December, the company Give access For developers and trusted testers, and wrap certain features into Google products, but this is a “general version”.
The model suite includes 2.0 Flash, which is called “the best schematic for high-frequency, high-frequency tasks with large capacity”; 2.0 Pro experiments, focusing primarily on coding performance; and 2.0 Flash-Lite, which Google sees as “The most cost-effective model to date”.
Gemini Flash’s developers have text, image and video inputs of 10 cents per million tokens, while Flash-Lite (its more cost-effective version) costs 0.75 cents.
The ongoing release is part of Google’s broader strategy to invest heavily in “AI Agents” as the AI arms race has intensified among tech giants and startups.
YuanAmazon, MicrosoftOpenAI and Anthropic are also heading towards proxy AI or models that can perform complex multi-step tasks on behalf of users, rather than having to browse them to each individual step.
“Over the past year, we have been investing in developing more proxy models, which means they can learn more about the world around them and think about multiple steps ahead,” Google wrote in a December December. and act on your behalf under your supervision.” Blog Postsadding that Gemini 2.0 has “new advances in multimodality, such as local image and audio output – and the use of local tools”, the “Model Family” will enable us to build new AI agents that bring us closer to our vision for universality assistant. ”
Humans, AmazonThe latter AI startup founded by former Olympic research executives is a major competitor to developing AI agents. In October, startups explain Its AI agents are able to use computers like humans to perform complex tasks. The startup says Anthropic’s computer usage features allow its technology to interpret what is on the computer’s screen, select buttons, enter text, navigate the website and perform tasks through any software and real-time Internet browsing.
Jared Kaplan, chief science officer of humans, told CNBC in an interview at the time that the tool “can use computers in essentially the same way.” He said it can accomplish tasks with “dozens or even hundreds of steps.”
Openai is released Similar tools Recently, a feature called “Operator” was introduced that will automate tasks such as planning a holiday, filling out forms, making restaurant reservations and ordering groceries. this Microsoft– The subsequent startup describes the operator as “a proxy that can go to the network to perform tasks for you.”
Earlier this week, Openai announced another tool Called in-depth research This allows AI agents to compile complex research reports and analyze user-selected questions and topics. Google launched a similar tool of the same name in December – Deep Dive – the tool is “Research Assistant, Explore complex topics and compile reports on your behalf”.
CNBC first report In December Google will introduce multiple AI capabilities in early 2025.
“In history, you didn’t always need the first place, but you have to perform well and really be the best product on the product,” CEO Sundar Pichai said at the strategy conference at the time. “I think That’s all for 2025.”
