Introduction
Cognition launched Devin in 2024 as "the first AI software engineer" and charged a high price to get in. The 2026 version is a different product. Devin Apps is now a tiered lineup: a desktop app with Tab completions and inline edits, cloud agents that work on their own and test in a browser, and always-on Devin Bots for set roles such as security review. Paid plans start at $20/mo, and there is a free plan.
This review covers what the vendor's pages say the product does, what the plans cost, and what independent reports have found about output. scored.tools did not run Devin on its own tasks. The scores below come from our rubric, built from published evidence. Devin Apps scores 5.8 out of 10 overall. It does well on ease of use and integrations and badly on output quality and problem fit.
| Criterion | Score | Basis |
|---|---|---|
| Output quality | 3 | Answer.AI's test, as reported by The Register |
| Ease of use | 8 | Free plan, CLI quickstart, first-session tutorial in the docs |
| Pricing value | 7 | Published plans at $20, $80 and $200 per month, extra usage at API pricing |
| API and integration | 8 | API docs, a CLI that controls cloud sessions, seven named integrations |
| Problem fit | 3 | ITPro testers reported fumbled tasks and salvage work |
Key Features
Autonomous agent, cloud or local
The main product is an agent that takes a task, writes code, tests the result in its own browser and opens a pull request. Per Cognition, it can run in the cloud (Devin Cloud) or on your own machine. Cloud agents need the Pro plan or higher. Running in the cloud means you can hand off a ticket and close your laptop, which is the point of the product.
Devin Bots
Devin Bots are always-on agents set up for a role. Cognition gives security review and design engineering as examples. They sit on the Max plan ($200/mo), the Teams plan and Enterprise. Pro users do not get them.
Desktop app with Tab and inline edits
The desktop app covers the same ground as Cursor and Windsurf: Tab completions and inline edits as you type. Both are unlimited on every plan, including Free, so the free tier works as an everyday editor even if you never touch the agent.
Models: SWE-2 plus frontier models
Cognition ships its own model, SWE-2, which Pro users can use for free. The pricing page also lists frontier models from OpenAI, Claude, Gemini and SpaceXAI, plus open-source models, and on Pro and Teams it lets you use an existing ChatGPT plan. The Free plan has limited model access.
Integrations and API
Devin connects to Slack, Teams, Linear, Jira, GitHub, GitLab and Bitbucket. Bitbucket and Jira support matters for shops that never moved to GitHub, which some rivals serve less well. The docs include an API section of 315 pages and a CLI that controls cloud sessions, so you can drive Devin from scripts and CI as well as from the app.
Pricing Breakdown
Cognition publishes its plans openly. The Teams line is the one to read closely.
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | Light agent quota, limited models, unlimited inline edits and Tab completions |
| Pro | $20/mo | Higher quotas, full model access, free SWE-2, Devin Cloud, extra usage at API pricing |
| Max | $200/mo | Everything in Pro, Devin Bots, much higher quotas |
| Teams | $80/mo plus $40/mo per full dev seat | Devin Bots, up to 100 users, multiplayer sessions, SAML/OIDC SSO, cost controls with 3 usage tiers, centralized billing |
| Enterprise | Custom | SCIM, role-based access control, dedicated deployment, account management, top-priority support |
A few things stand out. Per the pricing page, paid plans include a usage allowance that refreshes daily and weekly. Go past it and you buy extra usage at API pricing. Cognition says cost per message "varies based on the model used, the task size and complexity, and the reasoning required." So the base fee sets your minimum spend and tells you little about your maximum. An agent that loops on a hard task will cost more than one that finishes a simple one.
The Teams plan costs more than the headline figure. At $80/mo for the plan plus $40/mo per full dev seat, a five-person team pays $280/mo before any extra usage. The cost controls with three usage tiers are there for a reason. Set them before you invite anyone.
The $20 Pro entry price for cloud agents is low for this category. Bots, though, take a 10x jump to Max, and an individual who wants a standing security-review bot pays $200/mo.
Pros & Cons
Pros
- $20/mo for cloud agents is a low entry price for autonomous coding agents.
- The free tier has unlimited Tab completions and inline edits, so it works as an editor on its own.
- Wide git host and issue tracker support, including Bitbucket, GitLab and Jira.
- Enterprise options cover SSO, SCIM, role-based access and a dedicated deployment.
- The API docs and CLI make scripted, CI-driven use realistic.
Cons
- Devin Bots need the $200/mo Max plan, Teams or Enterprise.
- Extra usage is billed at API pricing, so real spend can run well past the base fee and is hard to forecast.
- The free agent quota is light and model choice is limited.
- The task results on Devin's homepage are vendor demos. The independent evidence on output is poor.
That last point drives the score. Answer.AI ran Devin on 20 tasks and, as reported by The Register, about 15% succeeded and 14 failed outright. The Register also noted that Cognition's own benchmark covered only a 25% sample. ITPro reported testers spending a lot of time rescuing failed attempts. Both reports cover earlier versions of Devin, before SWE-2 and the current lineup. We found no newer independent test, so they are the best public evidence available, and Cognition's claim of merge-ready PRs is unconfirmed by outside testing.
Who Is It For
Devin Apps suits a developer or small team with a steady supply of well-scoped tickets: dependency bumps, small bug fixes with a clear repro, test coverage, boilerplate endpoints. These are tasks you can check quickly and throw away cheaply if the agent gets them wrong. The Slack and Jira hooks fit teams that already file work as tickets and want an agent to pick some of them up.
It is a weaker fit if you want an agent to own large, vague features. The public evidence says that is where it fails, and failed runs still burn usage at API pricing. If you mainly want strong autocomplete and in-editor chat, Cursor, Windsurf and GitHub Copilot cover that ground. If you want an open-source agent you can self-host and inspect, look at OpenHands.
For procurement, Enterprise has SCIM, RBAC and a dedicated deployment option, which clears the usual security review checklist. Ask for usage caps in writing, because the per-message cost model is the hardest part to budget.
Verdict
Devin has grown from one expensive agent into a tiered product with a desktop app, cloud agents and bots, starting at $20/mo. The plans are clearly published, the free tier is a usable editor, and the integrations reach further than most. Against that, the only independent output tests on record are poor, and usage-based extras make the real monthly cost hard to predict.
Our recommendation: start on Free to try the editor, then pay for one month of Pro and give the cloud agent a batch of small, checkable tickets from your own backlog. Track how many PRs merge without rework and what extra usage costs you. Hold off on Max or a Teams rollout until those numbers justify it. At 5.8 out of 10, Devin Apps is worth a scoped trial. Do not commit a team to it on the strength of the vendor's demos.