Violently beating GPT-3.5, Google Gemini declares it free of charge! The most powerful code generation tool is launched to support 20+ languages

Google brings generous benefits to developers at the end of the year!

One week after the release of Google Gemini, the API for developers is finally online.

Currently, both Gemini Pro and Gemini Pro Vision can be tried for free.

Although it supports up to 60 requests per minute, it can basically meet the needs of most application development.

When it goes online early next year, Google will officially start charging.

API pricing will be consistent with GPT-3.5, i.e. USD 0.001 per thousand input tokens and USD 0.002 per thousand output tokens

Gemini Pro

##GPT-3.5

Specifically:

- Currently supports 32K context window, and will be further expanded in the future.

- Other basic functions are also supported: function calling, embedding, semantic retrieval, and custom knowledge.

This product supports 38 languages, covering more than 180 countries and regions around the world

- Supports API output of text, but the output is temporarily unavailable Only text is supported.

SDK support includes Python, Android (Kotlin), Node.js, Swift and JavaScript

In terms of fine-tuning, users can directly access the Gemini Pro API through Google AI Studio, and users can also fully customize Gemini through Vertex AI.

In this process, users can use their own data to build their own Gemini conversational robots without writing code. This robot supports multiple functions such as RAG models, search functions, embedding functions, and dialogue scripts

Google also promises that it will not use any user data and IP addresses to train models

In terms of capabilities, according to Google’s evaluation, Gemini Pro not only outperforms GPT-3.5 in most text benchmarks, And because it natively supports multi-modality, it directly implements dimensionality reduction in image evaluation.

The code generation tool is officially launched

On the same day, Google officially announced the full launch of Duet AI for Developers, a suite of AI-assisted tools for code completion and generation.

It’s very exciting to be blessed by Gemini models in the coming weeks

You must know that the most popular coding platform tool among developers today is none other than Microsoft's GitHub's Copilot.

In contrast, Google uses a data set from 25 companies' platforms to help developers build applications and troubleshoot code.

For example, Confluent, HashiCorp and MongoDB will provide data to train Duet AI for Developers to help developers write code for its platform.

#While data from these partners will mostly be useful in code completion and build experiences, Datadog, JetBrains, and LangChain will provide documentation and knowledge sources . After rewrite: While data from these partners will be useful for code completion and build experiences, Datadog, JetBrains, and LangChain will provide documentation and knowledge resources

In Duet AI for These documentation and knowledge resources are probably the most useful in Developers’ chat experience. For example, by using this data, the service can provide development and operations teams with information on how to create test automation, Information on troubleshooting issues and fixing bugs in production

How can artificial intelligence help get rid of things we don’t like to do, says Richard Seroter, one of Google Cloud’s evangelists. How do we make coding better? That’s a lot of what we’ve been chasing. How do we build an AI assistant that meets the tools developers use but adds a little bit of Google to it?

How to add Google elements to these tools?

For example, make sure models are trained on the latest cloud native practices and integrate them into the Google Cloud Console on top of all the popular integrated development environments developers already use.

Seroter emphasized that Google is focusing on the entire Duet AI product series, of which Duet AI in Security Operations is now officially released as an enterprise-level product.

The development process of Google’s AI coding tools is basically the same as that of every other technology company’s products.

#For example, Seroter pointed out that code auto-completion in IDEs that developers are already familiar with helps developers keep their workflow intact.

Google, like other major companies, believes that these tools do not replace coding skills, but help improve developer work efficiency

According to the latest data, artificial intelligence-driven technology services company Turing’s productivity increased by 33% after adopting Duet AI for Developers

Currently, Duet AI for Developers supports more than 20 languages , which includes C, C#, Java, JavaScript and Python Logging, and the integration of Smart Actions, which Google describes as one-click shortcuts for performing tasks such as unit test generation.

Before the end of January next year, Duet AI for Developers will be free and open. After that, Google will also adopt a charging system of $19 per month, on an annual basis.

The above is the detailed content of Violently beating GPT-3.5, Google Gemini declares it free of charge! The most powerful code generation tool is launched to support 20+ languages. For more information, please follow other related articles on the PHP Chinese website!