Devin 2.0 has arrived: Cognition reduces the price of an AI software engineer from $500 to $20 per monthly

Cognition AI, a San Francisco-based startup backed by Founders Fund (also known as Cognition labs), made a name out of nothing in early 2024 when it released Devin, an AI-powered software developer that could work with human developers and perform tasks autonomously using natural language instructions from a human dev via a prompt window or a separate third-party messaging app Slack.

However, the AI dev landscape is rapidly evolving, with many companies offering similar functionality, and autonomous or semiautonomous coding agent, including GitHub Copilot AWS Developer Q. Codeium’s Windsurf (19459056), and Cursor

Cognition has been using OpenAI’s GPT-4 series and GPT-4o since its inception to power Devin.

Today, Cognition has retaliated with Devin 2.0an updated version of their agent-native platform for software development. It is unclear what foundation model powers this version.

This new release is now available to the general public and includes a number of features that are designed to make collaboration between developers and Devin’s autonomous agent easier and more productive.

In an age of economic uncertainty, and Cognition has also provided a welcome relief by slashing its prices. Devin 2.0 can be purchased for as little as $20 per month(or $2.25 per “Agent Compute Unit”which is how Cognition measures compute resources needed to run Devin). Previous versions of the software were priced at around $500 per monthly.

What else does Devin 2.0 offer?

Parallel Devins, and a cloud-based IDE

Devin v2.0 builds on CognitionLabs’ earlier efforts to streamline the software development process by allowing users work alongside autonomous agents.

This latest version introduces a cloud-based interactive IDE environment that allows users to spin up several Devins simultaneously, effectively handling multiple tasks at once. The Devins can work independently, but users have the option to step in to edit, review, or guide progress at any time.

Interactive planning and task scoping are two key features of Devin 2.0. This feature allows developers with vague or incomplete ideas to collaborate with Devin and scope out a detailed plan.

Within a few seconds of starting the session, Devin analyzes the codebase and identifies relevant files. It then proposes an initial strategy, even without any specific instructions or guidance from the user.

The user can then review the plan and make any necessary adjustments to ensure alignment, before allowing Devin execute.

Exploring the codebase of your enterprise

Devin 2.0 introduces Devin Search as a tool to help users navigate and understand their codebases better.

The Search function allows developers to ask specific queries about their code and receive detailed responses with relevant code snippets. Users can activate Deep Mode for more complex queries requiring deeper exploration.

The new release also includes Devin Wiki – a feature which automatically indexes repositories once every few hours.

Devin Wiki creates comprehensive documentation, including architecture diagrams, links to source code, and other relevant information, providing developers with an organized and constantly updated reference.

Cognition Labs says Devin 2.0 is more efficient than ever.

The company claims that the latest version is able to complete over 83% more junior level development tasks per Agent Compute Unit compared to its predecessor. Beta users reported similar performance gains when testing.

Users are able to interact with Devin 2.0 via a VSCode inspired interface, which allows them to review and edit Devin’s code as well as run tests directly in the platform’s environment. This flexibility allows for both hands-on or hands-off workflows depending on the user’s preference.

Devin 1.2 gives you a boost

Early in 2025, the company released Devin version 1.2. This included enhancements focusing on in-context logic and voice command integration. These improvements allowed Devin to better analyse code repositories and recognize patterns. They also enabled it to reuse existing code where appropriate.

Users can also send voice messages to the agent, streamlining their interaction. Devin 1.2 introduced enterprise-oriented features, including machine snapshots that simplify login workflows as well as centralized admin controls to manage multiple Devin workspaces. Cognition Labs also shifted to a new usage-based billing system, which allows customers to pay extra for capacity beyond their subscription limits.

How Devin compares to other AI coding platforms and agents

Although Devin’s initial releases were hailed as a revolutionary solution for accelerating workflows, early feedback from users highlighted growing pains. Researchers and testers reported that the agent struggled with complex code, unnecessary abstractions and inconsistent task performance.

However, Devin was attracting interest from enterprise customers who wanted to integrate autonomous coding agents into the software development process.

The new capabilities and features of Devin 2.0, as well as the lower entry price, should be welcomed by developers, and could encourage them to adopt Cognition’s platform.

With GitHub Copilot’s Windsurf and Amazon Q Developer, among others, all offering free versions, Devin 2.0 is facing a tougher set of competitors in a hot market.