Cognition
Autonomous cloud software engineering agent with sessions, VMs, repository onboarding, review, automations, CLI, and desktop access.
Attributed output
GitHub public-attribution count, 2026-06-09
256KAttributed outputGitHub public-attribution count, 2026-06-09Commits and PRs on public GitHub provably produced by this agent. Exact if it is visible through its own bot or workflow, an estimate if measured through attribution it declares (these can swing 30 to 40 percent between measurement runs at multi-million scale), and a floor either way.Index rank · all agents
#6 of 40 CLI agents
#6Index rank · all agents#6 of 40 CLI agentsRanked by agent-opened PRs in the trailing 4 complete weeks, matching the index. Agents without a tracked PR series follow, ordered by total attributed output. Other filters or sorts change the list position.Business signal
Cognition annualized run-rate · May 2026
$492MBusiness signalCognition annualized run-rate · May 2026Reported by the press (May 2026), not officially disclosed or independently confirmed.Free / $20+ Pro
Entry price
Plan allowance and usage credits
Free / $20+ ProEntry pricePlan allowance and usage creditsNot published
GitHub stars
No public repo
Not publishedGitHub starsNo public repoNot published
npm demand/wk
Public npm counter
Not publishednpm demand/wkPublic npm counterWeekly package downloads as an install-demand floor; adjusted when the vendor's main install channel is no longer npm.From the public record
500 PRs · 201 repositories
Search window: – · captured Oct 14, 2026 (UTC)
Separate study: CodeRabbit follow-ups, May 25–October 11, 2026
In CodeRabbit’s May 25–October 11 audit, we found 2 original PRs crediting Devin, followed by 2 CodeRabbit auto-fix PRs; 1 follow-ups were merged. This cohort was selected through CodeRabbit, so it cannot tell us what fraction of this agent’s overall work receives follow-up fixes. Credits can name a model without identifying its client.
BerriAI/litellm: “fix(ui): move tags typed into key metadata JSON into the Tags field” (merged). OpenAgentsInc/openagents: “docs: spec for connecting a GPU door (GCE H100, Modal, local) to a chat” (merged).
This is a sample of publicly attributed work, not total usage or a market-share estimate. Repositories are not customers. Attribution is declared in metadata, not independently verified tool execution; it does not prove exclusive authorship or autonomy. PR status is recorded at capture; merged does not establish correctness.
Up to 100 most recently created search results in each of five consecutive 28-day windows; post-filtered for attribution in the PR body, bot author, or Codex branch. Not a random or exhaustive sample.
Returned records dated Jun 21, 2026–Oct 11, 2026.
Download the dated audit evidence (JSON)Weekly attributed output on public GitHub. Output visible through the agent's own bot or workflow is exact; output measured through declared attribution is an estimate. Window: 2025-02-03 to 2026-10-12.
Public GitHub commits attributed to Devin
Devin's work is visible through its integration; many commits inside its PRs land under the developer, so the pull-request count is the better task measure.
Last full week (ending 2026-10-18): 11,355 signed commits (-52.2% week over week), exact count.
Distinct users: 1,179 humans opened the 12,600 most recently sampled PRs, about 10.7 PRs per person.
Merge rate: 68.1% of its pull requests get merged (52,943 of 77,724 over 8 settled weeks, 2026-08-10 to 2026-10-04). The window ends about ten days back so recently-opened PRs have had time to merge.
GitHub review activity: 11.2% of its merged PRs had at least one GitHub review record before merging; the other 88.8% had none. This is a broad weekly signal and can include bot reviews; the human-review read below comes from the enriched PR sample.
In significant repos (company-owned or 10+ stars), 51.6% of its merged PRs got a human review. Across all sampled repos, the rate is 18.3%. Human review means a non-author, non-bot GitHub reviewer in the PR-lifecycle sample.
Median time to merge: 23 min. The unreviewed median is 0.33h. The human-reviewed median is 1.82h. From a 120-PR sample merged 2026-07-30 to 2026-08-28.
Test-inclusion: 45.8% of its sampled merged PRs touch a test file. This is a rough read on production readiness. 1 large PRs hit GitHub's 100-file path read cap, so treat this as a floor.
Commits per PR: 1.2 over the last 8 weeks. Both counts come from the same visible workflow, so this is a clean measure of how many commits make up one delegated task.
Flow share: 0.5% of the 1999 most recent public PRs sampled (10 PRs, 2 windows, last 2026-09-27).
Task lens: 161K agent-opened PRs. Devin's task unit is the pull request; private enterprise repositories are excluded.
Snapshot 2026-06-09, confidence High, tracked since 2026-06-09, +3.6% over 7 days. Read the methodology.
Share of Devin's own classified sampled work by segment; countries reflect where its PR authors are based. Full breakdowns: Languages, Frameworks, Industries, Map.
Pricing snapshot: 2026-09-25. Check the vendor for current terms.
Free, Pro ($20/month), Max ($200/month), and Teams ($80/month minimum) are available.
Task cost depends on the model, context, and tools used. A credit allowance is not a fixed token count.
| Plan | Price | Allowance | Unit | Token / Overage Notes |
|---|
Summary checked 2026-09-25.
Free, Pro ($20/month), Max ($200/month), and Teams ($80/month minimum) are available.
Devin is operated by Cognition, which also owns Windsurf; it shares Cognition's enterprise security posture.