Devin token usage
Devin does not expose raw LLM token counts on every plan. Self-serve accounts see quota and on-demand credits at Settings → Plans . Enterprise teams see ACUs (Agent Compute Units). Devin Desktop adds Plan Info token groups when your plan exposes them. Cloud sessions show per-session usage in Session Insights.
Hub: Tokenminning in Devin.
Does Devin show token usage?
| Plan / surface | What you see | Where |
|---|---|---|
| Self-serve (all surfaces) | Quota remaining, on-demand balance | Settings → Plans |
| Devin Desktop | Quota, on-demand $, ACUs, token groups (when available) | Settings → Plan Info |
| Devin Cloud | ACUs or quota per session | Session Insights, Settings → Consumption |
| Enterprise | ACUs by user/model | Get Consumption API |
Limitation: ACUs and quota bundle inference, actions, and some VM time — they are not a 1:1 map to provider-reported input/output tokens. Use token groups on Desktop when you need input/cache/output breakdown; use Session Insights categories on Cloud for action-level attribution.
Self-serve plans and quota
Cognition bills self-serve Devin through plan quota first, then on-demand credits. Quota is shared across cloud sessions, Devin CLI, and Devin Desktop on the same account.
| Plan | Price | Quota | Members |
|---|---|---|---|
| Free | $0 | Limited | 1 |
| Pro | $20/mo | Daily + weekly allowance | 1 (individual) |
| Max | $200/mo | Larger weekly allowance, no daily cap | 1 (individual) |
| Teams | $80/mo minimum | Per full seat: Pro-equivalent quota | Unlimited |
On-demand credits fund usage past quota. They roll over month to month and never expire. Teams plans share one credit pool across all members. Flex seats (free, no Desktop) draw only from that shared pool.
Teams billing mechanics:
| Full seats | On-demand credits in minimum | Monthly total |
|---|---|---|
| 0 | $80 | $80 |
| 1 | $40 | $80 |
| 2+ | $0 | $40 × seat count |
Full seats ($40/mo each) include Pro-equivalent quota and Devin Desktop access. Flex seats are for occasional cloud-only users.
Legacy ACU-based self-serve plans migrated to quota; on-demand credits keep the same dollar value as the ACUs you were used to. Details: Self-serve plans .
Where to read the meter
| Surface | Location | Units |
|---|---|---|
| Self-serve (all surfaces) | Settings → Plans | Quota remaining, on-demand balance |
| Devin Desktop | Settings → Plan Info | Quota, on-demand $, ACUs, legacy credits |
| Self-serve billing | manage your plan | Invoice, cycle reset |
| Enterprise desktop | Get Consumption API | ACUs by user/model |
| Devin Cloud | Session Insights, Settings → Consumption | ACUs or quota per session |
ACUs vs credits vs quota
| Unit | Meaning |
|---|---|
| Quota | Included allowance that refreshes on a calendar schedule (daily + weekly on Pro/Teams full seats; weekly only on Max) |
| On-demand credits | Prepaid usage past quota; shared on Teams; powers flex seats entirely |
| ACUs | Agent Compute Units — inference + actions + some VM time (cloud + enterprise) |
| Legacy credits | Often one credit per user message on premium Cascade models, regardless of tool calls in that turn |
Read the column your Plan Info or Plans screen actually shows — do not assume ACUs on a credit or quota plan.
What counts toward usage (cloud)
Per official usage docs , usage accrues from work Devin performs:
- Number and complexity of actions (planning, context gathering, execution, browser, shell)
- Virtual machine time and networking (typically a small fraction)
Devin does not consume usage while sleeping, waiting for your reply, waiting for tests, or during repo clone/setup (aside from minimal VM keep-alive). Windows sessions cost roughly 9% more than equivalent Linux sessions.
Token groups (desktop)
When your plan exposes breakdowns:
| Group | High value means |
|---|---|
| Input | Rules, MCP, index bloat |
| Cache write | Large first message or tool results |
| Cache read | Thread too long — new conversation |
| Output | Thinking models, verbose diffs, loops |
Adaptive self-serve rates (through July 7, 2026): ~2.00/M output, $0.10/M cache read — routes simple work to cheaper models underneath.
Enterprise APIs
Devin auth token for metering (not your Git PAT):
- Desktop: Get Consumption API , Cascade Analytics API
- Cloud: daily consumption , session metrics by category
Rotate API keys on the same schedule as CI secrets.
Cloud vs desktop vs CLI
| Workload | Meter surface | Guide |
|---|---|---|
| Autonomous multi-step PR | Cloud quota / ACUs | Cloud ACUs |
| IDE pair programming | Desktop Plan Info | Devin Local |
| Terminal quick fix | Same plan quota | CLI |
Desktop inference bills like other IDE agents. Cloud adds VM time, browser actions, and exploration overhead — meter separately.
After a heavy week
- Plans or Session Insights — which surface moved?
- Token group, quota burn, or ACU category — input vs output?
- Open the matching spoke: rules, MCP, instructing, cloud.