Google runs Gemini 2.0 Pro and Flash-Lite and Modeling Model thinking about YouTube, maps and search!

GEMINI series of Google from the large language models of artificial intelligence (LLMS) It started nearly a year ago with some embarrassing incidents to generate photosBut it has improved steadily, and it seems that the company is determined to make the second generation – GIMINI 2.0 – the largest and best for consumers and institutions.

Today, the company Declare The overall version of Gemini 2.0 Flash, Gemini 2.0 Flash-Lite, and an experimental version of Gemini 2.0 Pro.

These models, designed to support developers and companies, can now reach Google Ai Studio and Vertex AI, with Flash-Lite in Public Preview and Pro for Early Test.

“All these models will display multimedia inputs with the text out of the version, with more methods ready for general availability in the coming months,” wrote Koray Cavoogooglu, a chief technology in Google DeepMind. Blog advertisement – View a feature that Google brings to the table even as competitors like Dibsic and Openai Continue to launch strong competitors.

Google plays on its multimedia episodes

No Deepseek R1 and no New O3-Mini Model from Openai The multimedia inputs, i.e. images and download files or attachments.

Although Deepseek R1 can accept them on its website and chat mobile applications, but it leads to visual letters (OCR) technology more than 60 years old, to extract the text only from these downloads – no understanding or analysis of any of the features The other in it.

However, both are a new category of “thinking” models that deliberately take more time to think through answers and think about “idea chains” and the validity of their responses. This is opposed to typical LLMS such as the Gemini 2.0 Pro series, so the comparison between Gemini 2.0, Deepseek R1 and Openai O3 is a little bit of apples to software.

But there was some news on the front of the mind today from Google, too: Google Sundar Pichai CEO has moved to the social network x To announce that Google Gemini The iOS and Android mobile phone app has been updated using Gemini 2.0 Flash Model from Google Gemini 2.0, and it can be the model connected to Google’s current Google, YouTube, and Google Search, allowing a full new set of AI-AI-research And the interactions that can be compatible with them through services without such services as Deepseek and Openai.

I tried it shortly on the Google Gemini iOS app on my iPhone while writing this piece, and it was impressive based on my initial inquiries, thinking about the common denominators of the 10 best videos on YouTube most popular in the last month and also providing me from nearby doctors ’offices and opening hours /Close, all within seconds.

Gemini 2.0 Flash enters the general version

Gemini 2.0 Flash model, It was originally launched as an experimental version in DecemberNow ready for production.

It is designed for highly efficient artificial intelligence applications, provides low total responses and supports multimedia thinking on a large scale.

One of the main benefits is the competition in its context window, or the number of symbols that the user can add in the form of a claim and receives again in one reaction with monitoring with LLM or the application programming interface.

While many pioneering models such as the new O3-MINI from Openai, which first appeared in the past week, support only 200,000 or less than the distinctive symbols-about the equivalent of a 400-500-pages of information from the Information density-Gemini 2.0 Flash supports a million, which means It is able to deal with wide amounts of information, which makes it especially useful for high -frequency tasks and wide materials.

Gemini 2.0 Flash-Lite to bend the cost curve to the lowest level so far

GEMINI 2.0 Flash-Lite, at the same time, is a completely large new language model aimed at providing an effective AI solution without prejudice to quality.

Google DeepMind states that Flash-Lite outperforms its full predecessor (larger than the parameter), Gemini 1.5 Flash, on third-party standards such as MMLU Pro (77.6 % compared to 67.3 %) and SQL Bird (57.4 % against. 45.6 %), while preserving On the same pricing and speed.

It also supports multimedia input and features a context frame of one million symbols, similar to the full flash model.

Currently, Flash-Lite is available in the general inspection through Google Ai Studio and Vertex AI, with an expected year in the coming weeks.

As shown in the table below, Gemini 2.0 Flash-Lite is priced at $ 0.075 per million code (input) and $ 0.30 per million codes (output). Flash-Lite is placed as an option at very reasonable prices for developers, outperforming GEINI 1.5 across most criteria while maintaining the same cost structure.

Logan Kilpatrick has highlighted the ability to bear costs and the value of the models, saying: “GEINI 2.0 Flash is the best value for any LLM, it’s time to build!”

In fact, compared to the other traditional LLMS available via the provider applications interface such as Openai 4o-Mini (0.15 dollars/0.6 dollars per million code in/outside), Claude Al -Bushra (0.8 dollars/4 dollars! Per 1 meter in/outside), and even the traditional Deepseek LLM V3 ($ 0.14/0.28 dollars), at Gemini 2.0 Winf Flash is the best bang for p.

Gemini 2.0 Pro reaches an experimental availability with a symbolic context window 2 million

For users who need more advanced capabilities of artificial intelligence, the Gemini 2.0 Pro (experimental) model is now available.

Google DeepMind describes this as the most powerful performance manufacturer and dealing with complex claims. It is characterized by a window of 2 million people and improved thinking capabilities, with the ability to integrate external tools such as Google Search and the implementation of software instructions.

Sam Witteveen, co -founder and CEO of Red Dragon AI and an external Google developer expert for automatic learning who often shares Venturebeat, Discuss the professional model in YouTube review. “The new Gemini 2.0 Pro model has a window of millions of millions, supports tools, implementing code, job invitation, swinging with Google Search-everything we had in Pro 1.5 but was improved.”

He also pointed to the Google Repeating approach to the development of artificial intelligence: “One of the main differences in the Google strategy is that it issues experimental versions of models before they reach GA (can be accessed in general), allowing rapid repetition based on comments.”

Performance standards also show the capabilities of the Gemini 2.0 Family Family. Gemini 2.0 Pro, for example, outperforms the flash and flash through tasks such as thinking, multi -language understanding and tall context.

Safety of artificial intelligence and future developments

Besides these updates, Google Deepmind is implementing new safety and safety measures for Gemini 2.0 models. The company takes advantage of reinforcement learning techniques to improve response accuracy, using artificial intelligence to criticize and refine its own outputs. In addition, an automatic security test is used to determine weaknesses, including indirect injection threats.

In the future, Google DeepMind plans to expand the capabilities of the Gemini 2.0 Model family, with additional methods that exceed the text in general in the coming months.

Through these updates, Google enhances its promotion to develop artificial intelligence, as it offers a set of models designed for efficiency, the ability to bear costs, solve advanced problems, and respond to the climb of Deepseek with a set of their own models that range from strong to very strong and exciting Very and very strong and elaborate at a reasonable prices a little less (but still largely) at reasonable prices.

Will this be enough to help Google eating in some of the AI ​​Enterprise market, which was once dominated by Openai and is now by Deepseek? We will continue to track and teach you!

Daily visions about business use cases with VB daily

If you want to persuade your boss at work, you have covered VB Daily. We give you the internal journalistic precedence over what companies do with obstetric artificial intelligence, from organizational transformations to practical publishing operations, so that you can share visions of the maximum return on investment.

Read with us privacy policy

Thanks for subscribing. Check more VB news bulletins here.

An error occurred.