Google has just released what could be the most powerful AI coding assistant, and it is free to download and use for everyone. Called Gemini Code Assist, the AI assistant is based on a specialized version of Google’s flagship large language model (LLM) and it has a lot of great features and capabilities.
Gemini Code Assist comes as the competition for commercial AI heats up between frontier AI labs and tech giants. In the past week, we’ve also seen the release of Claude 3.7 Sonnet, which includes a coding assistant.
What is Gemini Code Assist
Gemini Code Assist is a product that is powered by Gemini 2.0. According to Google, it is a fine-tuned version of the model that has been trained on coding data gathered from “a large number of real-world coding use cases.” There is not much detail on the model’s architecture or the training data composition, and since the model is private, we can only speculate.
But given the free quotas that Google is providing and the speed at which the model responds, we can assume that behind the scenes, it is using Gemini 2.0 Flash Thinking, the smaller version of the Google LLM family (which by the way is an awesome model and my first option for most tasks).
Google Code Assist has a context window of 128,000 tokens, making it very useful for tasks that span several files (another sign that it might be a derivative of Gemini 2.0 Flash Thinking, which has a similar context window).
(The full Gemini models support up to 2 million tokens. My guess is that coding and reasoning tasks are very hard to perform on very large token sequences because they require grabbing context from various parts of the context window, which is why reasoning models are still in the hundreds of thousands of tokens.)
How can you access Gemini Code Assist?
Google has already deployed its AI assistant across its vast empire of online apps, including developer tools such as Colab, Android Studio, and Firebase.
The new release brings Gemini Code Assist to a variety of integrated development environments (IDEs), including Visual Studio Code and JetBrains applications such as PyCharm. To get access to Gemini Code Assist, you only need to install the extension on your IDE and log in to your Google Account. The free version provides you with 6,000 requests per day and 180,000 per month (more than most developers would need and more generous than the quota on Gemini 2.0 Flash Thinking).
In the IDE, you can use Gemini Code Assist in different ways, including code generation and explanation. For example, you can select a code snippet and ask Gemini Code Assist to explain it. Or you can select a function and have the model generate unit tests for it. Gemini Code Assist will also generate smart autocomplete suggestions as you write code, which take into account the file’s context and your comments, making the coding experience much more pleasant.
Google has also released Gemini Code Assist for GitHub, which can help in reviewing code, applying changes, handling PRs, and managing repositories.
Along with the free version, Google has also released Gemini Code Assist Standard and Enterprise for power users and organizations.
I have become a huge fan of the Gemini models and I think that in the past few months, Google has managed to bridge much of the gap with OpenAI and Anthropic. With its vast distribution, data, and financial resources, I think that Google is likely to take the lead in generative AI tools in the near future.