AI cost · Cognition
Devin cost: what you actually pay
Self-serve plans ($0/$20/$200) run on an undisclosed token-based quota that resets daily/weekly; once it's gone, overage bills straight through at raw frontier-model API rates, and Enterprise still hides behind custom-quoted Agent Compute Unit (ACU) pricing.
Public pricing
| Plan | Price | Billing | Includes |
|---|---|---|---|
| Free | Free | monthly | Light daily/weekly quota, limited model availability, unlimited inline edits and Tab completions, Devin Review + DeepWiki |
| Pro | $20 / moGrandfathered legacy users reportedly retained a $15/mo rate | monthly | Increased quota, full frontier model access (OpenAI, Claude, Gemini), free SWE-1.7 and OSS models, Devin Cloud agents, can buy extra usage at API pricing |
| Max | $200 / mo | monthly | Significantly higher quota than Pro, same feature set |
| Teams | $40 / seat/moPlus an additional $80/mo base platform fee (effectively $80 + $40 x seats); legacy accounts reportedly retained ~$30/seat | monthly | Unlimited team members, shared usage pool, centralized billing, admin analytics dashboard, priority support |
| Enterprise | CustomReportedly still uses ACU-style metering (historically ~$2.25/ACU on legacy Core plans) but current rates are custom-quoted and unpublished | custom | SAML/OIDC SSO, dedicated account management, centralized admin controls, dedicated deployment option |
Self-serve tiers meter an undisclosed token-based 'quota' refreshing daily and weekly; Cognition does not publish the conversion rate. Historically (pre-2026 restructure) Devin billed in Agent Compute Units (ACU) at ~$2.25/ACU (Core) or ~$2.00/ACU (Team), where 1 ACU approximates 15 minutes of active agent work; enterprise contracts reportedly still use ACU-style metering today at custom-quoted rates. Once self-serve quota is exhausted, Pro/Max/Teams users can buy extra usage billed at the underlying model's raw API rate with no bundled discount.
Last updated: 2026-07-22. Sourced from Devin Pricing (official) ↗, eesel AI — Cognition AI pricing explained ↗.
Where the real Devin cost hides
The sticker is the smallest part of the story. For Devin, these are the line items that quietly inflate the bill:
- Multi-hour or ambiguous multi-file tasks consume quota non-linearly — a stuck agent that retries and self-corrects can burn an outsized share of quota on one unresolved bug
- Overage is billed at raw frontier-model API rates with no discount, so a $20/mo plan can turn into a much larger bill mid-task with no warning ceiling
- Teams pricing is base-fee-plus-seat ($80 + $40/seat), easy to underestimate versus the $20/mo individual Pro sticker price
- Enterprise ACU pricing is custom-quoted and unpublished, so per-task cost is unknowable until under contract
What an audit finds on Devin
Because Devin runs autonomously over long horizons, an ambiguous ticket can trigger many build-test-fail-retry self-correction cycles that each consume compute; teams report the monthly sticker price is dwarfed by a handful of runaway multi-hour sessions that spill into API-rate overage.
Devin pricing FAQ
What's an ACU and am I still billed in them?
An Agent Compute Unit (ACU) normalizes VM time, model inference, and networking into one unit (~15 minutes of Devin work). Self-serve Free/Pro/Max/Teams plans now use a simpler daily/weekly quota, but Enterprise contracts reportedly still use ACU-style metering at custom-quoted rates.
What happens when quota runs out?
Free users wait for the next reset; Pro/Max/Teams users can purchase extra usage billed at the underlying model's raw API token rate.
What is Devin really costing you?
Spendassay connects your Devin spend to actual usage and engineering output — a proof-level AI cost report with the wasted dollars named. Neutral across every AI vendor.
Free · read-only · no card
Other autonomous agents pricing
More: the complete AI cost management guide · every tool, with prices · cost per outcome · compare platforms · how we measure