Visual China Group | Getty Image
Google On Wednesday, Gemini 2.0 (the “most capable” artificial intelligence model suite) was released to everyone.
According to Google, in December, we accessed a reliable tester with the developer and wrapped some features in Google products, which is a “general release.”
Model suite contains 2.0 flashes, which are claimed as “the main model for large large-scale high-frequency tasks” and 2.0 Pro Experience for coding performance, 2.0 Flash-lite, 2.0 Flash-Lite, 2.0 Flash-Lite. But there is the company. It is called “the most expensive model so far.”
Gemini Flash costs 1 million tokens for developers, images, and video input, but the FLASH-LITE, a more cost-effective version, costs 0.75 cents. Masu. The token refers to the individual unit of the data processed by the model.
The continuous release is a wide range of strategies in which AI Arms Race makes a large amount of investment in AI agents as the AI Arms race gets hot between the high -tech jirates and startups.
Meta,Amazon, MicrosoftOpenai and humanity are not agent AI or users to take all individual steps, but to models that can complete complex multiple tasks on behalf of users.
“Last year we have invested in the development of more agent models. In other words, they understand the world around you, think one step ahead, and take action on your behalf. You can do it, “Google wrote on the December blog. If you post, GEMINI 2.0 has a “new progress of multi -modernity, such as the use of native tools, such as native image and audio output”, and the model family can “build a new AI agent that we can approach our vision. You can do it.
Amazon, an AI startup supported by EX-OPENAI research executives, is a major competition for developing AI agents. In October, Anthropic stated that AI agents could use human -like computers to complete complex tasks. With anthropic computer use function, the technology interpretes the one on the computer screen, selects the button, enters the text, navigates the website, and can execute the task via the software and real -time Internet browsing. Masu.
JARED KAPLAN, the highest scientific officer of mankind, tells CNBC in an interview at the time, “basically using a computer like us”. He stated that he could do tasks in “dozens or hundreds of steps”.
Openai has released a similar function called an operator that automates tasks such as vacation planning, form forming, restaurant reservations, and food orders. The startup supported by Microsoft explained the operator as “an agent to go to the web to run the task for you.”
Earlier this week, Openai introduced Deep Research. This allows AI agents to edit complex research reports and analyze questions and topics for user selection. In December, Google launched a similar tool (Deep Research) with the same name, “Exploring the research assistant, exploring complicated topics and editing a report on behalf of you).
CNBC first reported in December that Google would introduce some AI functions in early 2025.
“You don’t need to be the first in history, but you need to be the best in your class as a product,” said CEO’s Sundar Pichai at the strategic conference at the time. “I think it’s all of 2025.”
