Cognition Devin review

4.4/5

AI coding agent

Devin is Cognition's autonomous AI software engineer. You assign it a task or ticket (via web, Slack, Linear/Jira, or API) and it plans, writes code, runs commands in its own cloud VM, and opens pull requests largely without supervision. I

Visit Cognition Devin → last verified 2026-05-31

Affiliate link · how we make money

Best for autonomous coding. Cognition Devin starts $20/mo (Pro) + usage overage on a base subscription + usage credits model. Pro is $20/mo and includes a daily/weekly Devin usage quota plus a Windsurf IDE quota; once you exhaust it, extra work is billed pay-as-you-go via prepaid on-demand credits priced in dollars, equivalent to the old ACU rate. One ACU is roughly 15 minutes of active Devin work and has historically run about $2.25 each, so a busy hour of autonomous engineering is around 4 ACUs, near $9/hour on top of the base fee. Max ($200/mo) and Teams ($80/mo org) just raise the included quota before the same per-credit overage kicks in. Real monthly spend therefore tracks how many active hours you actually run Devin, not the sticker price.

Spec

pricing modelbase subscription + usage credits
entry price$20/mo (Pro) + usage overageverify livedecode →
free tierYes. $0/mo plan with limited Devin usage, plus Devin Review and DeepWikisource →
self-hostNo
MCPBoth
integrationsNative: GitHub, Slack (kick off runs by tagging @Devin), Jira, Linear, Secrets Msource →
API accessYes
affiliateno public program

The gotcha

the pricing catch

The $20/$80 sticker price is just a base subscription with an included usage quota. The real cost is consumption-based: once you exhaust the quota, every additional chunk of active work is billed pay-as-you-go (the ACU model, ~$2.25/ACU ≈ $9/active hour historically). Auto-reload of credits is on by default so spend can run silently past the base fee. ACUs are consumed by the active work Devin does, not by wall-clock or idle time (it auto-sleeps when waiting), so a session that sits idle is cheap but a busy one stacks credits fast.

Decode the real cost →

Best for

  • You want a fully autonomous AI engineer that can pick up a Jira/Linear ticket or Slack request, work in its own cloud environment, and open a finished pull request with minimal hand-holding, and you can manage usage-based billing that adds up on top of the base subscription.

source: devin.ai/pricing →

Cognition Devin pricing tiers

Prices move. Cells flagged verify link to the live vendor page.

Decode Cognition Devin pricing →
tiermonthlyannualincludedunit
Free $0 n/a Limited Devin usage; Devin Review; DeepWiki per user
Pro $20 n/a Monthly Devin usage quota + Windsurf IDE usage quota; pay-as-you-go overage past quota; Slack, Linear, MCP integrations; 1 member per user/mo + usage overage
Max $200 n/a Increased Devin usage quota + increased Windsurf IDE quota; pay-as-you-go overage; 1 member per user/mo + usage overage
Teams $80verify n/a Unlimited team members, shared/collaborative access, centralized billing, admin dashboard with analytics, unlimited concurrent sessions, Devin API; included usage quota + pay-as-you-go overage per org/mo (from $80) + usage overage
Enterprise Custom n/a SAML/OIDC SSO, centralized admin controls, dedicated account team, self-hosted source control/artifact repo support, custom ACU rate set in order form, custom terms custom

// pools and per-unit rates are volatile · cells flagged verify link to the live vendor page

Compare Cognition Devin with…

Questions

How much does Cognition Devin cost?

Cognition Devin starts from $20/mo (Pro) + usage overage. Pro is $20/mo and includes a daily/weekly Devin usage quota plus a Windsurf IDE quota; once you exhaust it, extra work is billed pay-as-you-go via prepaid on-demand credits priced in dollars, equivalent to the old ACU rate. One ACU is roughly 15 minutes of active Devin work and has historically run about $2.25 each, so a busy hour of autonomous engineering is around 4 ACUs, near $9/hour on top of the base fee. Max ($200/mo) and Teams ($80/mo org) just raise the included quota before the same per-credit overage kicks in. Real monthly spend therefore tracks how many active hours you actually run Devin, not the sticker price.

Can you self-host Cognition Devin?

No, Cognition Devin is cloud-only.

Does Cognition Devin support MCP?

Cognition Devin supports MCP as both client and server, so it works with assistants like Claude and ChatGPT.

What's the catch with Cognition Devin's pricing?

The $20/$80 sticker price is just a base subscription with an included usage quota. The real cost is consumption-based: once you exhaust the quota, every additional chunk of active work is billed pay-as-you-go (the ACU model, ~$2.25/ACU ≈ $9/active hour historically). Auto-reload of credits is on by default so spend can run silently past the base fee. ACUs are consumed by the active work Devin does, not by wall-clock or idle time (it auto-sleeps when waiting), so a session that sits idle is cheap but a busy one stacks credits fast.