# Discover Devin's Pricing and Plans for 2026 Discover Devin’s pricing and plans for 2026. Compare Core, Team, and Enterprise tiers, costs, and features to find the right fit. ![](https://framerusercontent.com/images/GjPJ8lgQ2s9KH4YirhymwwZxVY.png?width=1152&height=1152) Pensero Pensero Marketing Apr 21, 2026 Devin by Cognition AI moved from early access to general availability with a pricing model built around a concept most engineering tools don't use: compute units rather than seats. Understanding how that model works, what each plan actually gives you, and how to calculate whether the investment is justified is not straightforward. This guide breaks it down. ## **How Devin Pricing Works** Devin does not charge per user or per seat in the traditional sense. It charges based on consumption of ACUs, Agentic Computing Units, which are Cognition's normalized measure of the resources Devin uses while actively working: virtual machine time, model inference, and networking bandwidth. Seats are unlimited across all plans. What you are buying is working time, not access licenses. This model reflects the nature of autonomous agent work more honestly than per-seat pricing would. A tool that can run multiple sessions in parallel, work overnight, and process large codebases is not well-served by a model that charges the same whether it runs for five minutes or five hours. ## **Devin Plans in 2026** ### **Core** The Core plan is Devin's pay-as-you-go entry point, starting at $20. ACUs are billed at $2.25 each, and auto-reload settings can be configured so teams are never blocked waiting for credits. Up to 10 concurrent Devin sessions are supported, and the plan includes unlimited users, collaborative access, the Devin IDE, Ask Devin conversational interface, Devin Wiki, Devin API, and the Advanced Capabilities layer that allows Devin to orchestrate managed Devins in parallel, analyze past sessions, create and improve playbooks, and manage a knowledge base. There is no monthly commitment. Core is best suited for individual developers or small teams with variable usage patterns who want access to Devin's full capability set without a fixed monthly spend. ### **Team** The Team plan is $500 per month and includes 250 ACUs at $2.00 per ACU, a slight discount from the Core rate. Concurrent sessions are unlimited on this plan, removing the 10-session cap of Core. The Team plan also provides access to early feature releases and research previews, and includes an optional onboarding call with the Cognition team. Auto-reload settings are available for on-demand consumption beyond the included ACU allocation. For engineering teams running Devin regularly across multiple workstreams, the Team plan's unlimited concurrency is the meaningful differentiator. The ability to spin up parallel Devin sessions is what made Nubank's large-scale [ETL migration](https://www.forbes.com/councils/forbestechcouncil/2023/05/08/etl-the-overlooked-hurdle-in-successful-cloud-migration/) possible: rather than working sequentially through over 100,000 data class implementations, they delegated subtasks to multiple Devin instances running simultaneously, collapsing what would have been an 18-month project into weeks. ### **Enterprise** The Enterprise plan is custom-priced and built for organizations that need Devin deployed in their own infrastructure with full security controls. Key additions over Team include deployment within a virtual private cloud, SAML and OIDC SSO, centralized enterprise admin controls, teamspace isolation for different departments, a dedicated account team, custom terms, and centralized billing and usage analytics across multiple Devin organizations. The Enterprise version of Devin, Devin Enterprise, is described by Cognition as the most capable version of the platform. Enterprise is the relevant plan for organizations in regulated industries, those with strict data residency requirements, or large organizations running Devin across multiple teams who need consolidated governance. ## **Understanding ACUs in Practice** One ACU represents approximately 15 minutes of active autonomous work by Devin. The Team plan's 250 ACUs translate to roughly 62.5 hours of autonomous engineering time per month. ACU consumption is not constant across tasks. Variables that affect consumption include task complexity, the specificity and quality of the prompt, the size of the codebase Devin is navigating, the number of files it touches, and session runtime. A well-scoped, specific prompt on a bounded codebase will consume fewer ACUs than an open-ended task on a large monorepo. Devin only consumes ACUs when actively working or when the virtual machine is running, not during idle time. API usage does not carry additional charges beyond the ACUs consumed by the sessions started programmatically. Teams on Team or Enterprise plans with API access pay only for the agent compute used. All code generated by Devin is owned by the customer regardless of plan. Inputs and outputs are treated as the customer's intellectual property. ## **The ROI Case: What Nubank's Migration Shows** The most documented case study for Devin's ROI is Nubank's core ETL migration. The project involved moving over 6 million lines of code across a monolithic repository into sub-modules, with more than 100,000 data class implementations requiring individual migration. The expected scope was 18 months of work distributed across over 1,000 engineers. With Devin handling the repetitive migration subtasks autonomously, Nubank achieved 8 to 12x efficiency gains measured in engineering hours, and over 20x cost savings on the delegated scope compared to human engineering time. Data, Collections, and Risk business units completed their migrations in weeks rather than months. Fine-tuning Devin on examples of prior manual migrations doubled task completion scores and reduced per-subtask time from roughly 40 minutes to 10. The economics of this specific case were driven by three factors: the high volume of repetitive, discretionary subtasks, the ability to run multiple Devin instances in parallel, and the relatively low cost of ACUs compared to fully-loaded engineer hourly rates. Not every engineering workflow has these characteristics. Large-scale migrations, dependency updates, test generation, and documentation tasks tend to produce the best ROI. Open-ended feature development in complex, ambiguous problem domains tends to require more human involvement and produces less predictable ACU efficiency. ## **What Devin Does Not Tell You** Devin measures its own usage. It does not measure whether that usage is making your engineering organization more competitive, whether quality is holding, whether the teams using it are outperforming those that are not, or whether the investment is justified relative to what comparable organizations are achieving. Those questions require a different kind of platform. [Pensero](https://pensero.ai/) measures AI coding tool impact at the work-item level, identifying AI-generated versus human-authored code across tools including Devin, Copilot, Cursor, and Claude Code. Pensero Calibrate lets leaders compare AI-adopter and non-adopter cohorts on 11 complexity-weighted metrics with the industry median as a reference line. Pensero Benchmark ranks the organization against real anonymized production data from comparable organizations across 10 performance dimensions. For engineering leaders who need to answer the board's question, which is not "how many ACUs did we consume?" but "is AI making us more competitive?", Pensero is the measurement layer that sits alongside whichever coding tools the team adopts. Pensero pricing as of April 2026: free tier up to 10 engineers and 1 repository; $50/month premium; custom enterprise pricing. ## **Frequently Asked Questions** ### **How much does Devin cost in 2026?** Devin's Core plan starts at $20 with pay-as-you-go ACU billing at $2.25 per ACU. The Team plan is $500 per month and includes 250 ACUs at $2.00 each. Enterprise pricing is custom. Seats are unlimited across all plans. ### **What is an ACU?** An ACU, or Agentic Computing Unit, is Devin's normalized measure of the resources used while actively working on a task: virtual machine time, model inference, and networking bandwidth. One ACU represents approximately 15 minutes of active autonomous work. ### **How many ACUs does a typical task use?** It depends on task complexity, prompt specificity, codebase size, and session runtime. Well-scoped, specific tasks on bounded codebases consume fewer ACUs. Open-ended tasks on large monorepos consume more. Devin only charges ACUs when actively working or when the virtual machine is running. ### **What is the difference between the Core and Team plans?** The main practical differences are concurrent sessions (up to 10 on Core, unlimited on Team), the monthly ACU inclusion (pay-as-you-go on Core, 250 included on Team at a slightly lower per-ACU rate), and access to early feature releases and research previews on Team. Both plans include unlimited users. ### **Who owns the code Devin generates?** The customer owns all inputs and outputs regardless of plan. Cognition treats all generated code as the customer's intellectual property. ### **Is Devin worth the cost?** For high-volume, repetitive engineering tasks such as large-scale migrations, dependency updates, and test generation, the ACU economics can be compelling. Nubank's case demonstrates what is possible at scale with the right task type. For complex, ambiguous feature development, the ROI depends heavily on how well tasks are scoped and how much human review time the output requires. The honest answer is that the ROI calculation requires measuring delivery outcomes, not just ACU consumption, and that requires a measurement platform like Pensero rather than Devin's own usage dashboards. ### **Does Devin work alongside other AI coding tools?** Yes. Devin can be used alongside tools like Copilot, Cursor, and Claude Code, with different tools suited to different task types and oversight preferences. Measuring the combined impact of multiple AI tools on delivery and quality requires a platform that spans all of them at the work-item level, which Pensero provides.