AI Observability: Claude Code - Subscription and Billing

The AI Observability: Claude Code - Subscription and Billing dashboard shows how an organization uses its paid Claude subscription. It shows how many seats are active or idle, how much of their cost has been consumed, which members reach subscription limits, and which seats should be reviewed. The final section reconciles estimated costs with an equivalent of a subscription spend report.

Useful for: Claude subscription administrators, financial analysts, and development managers.

Data sources: gen_ai_metrics*, gen_ai_events*, and the claude_code_members, claude_code_seat_pricing, and claude_code_spend_report lookups.

---

Main Sections

1. Seats

Shows paid seats and their usage:

• seat costs for the period and the numbers of paid, active, and idle seats
• seat utilization: the proportion of seats that were used
• paid and active seats by day
• seats by type: active and idle

The seat price comes from the claude_code_seat_pricing lookup according to the member's seat type.

2. Budget and Consumption

Shows how much of the paid seat budget has been consumed through work in Claude Code:

• consumption for the period and per seat, utilization, and remaining seat budget
• seat payments and consumption, both cumulative and by month
• monthly budget utilization
• seat utilization by member and the distribution of members across utilization levels

The seat budget is the monthly seat price prorated to the days in the period. A negative remaining budget means consumption exceeds the seat cost.

3. Tokens

Compares token consumption with the seats' token budget:

• tokens used, the seat token budget, and its utilization
• effective price per 1M tokens
• tokens by type and day, and cache-read share by member
• summary by seat type: seat count, payment, consumption, tokens, and utilization

4. Errors and Rejections

Shows API errors and limit-related rejections:

• counts of API errors, limit-related rejections (429), members with rejections, and requests that exhausted their retry attempts
• API errors and requests that exhausted their retry attempts by day
• a table of rejections and errors by member, including the latest error

5. Seat Recommendations

Shows which seats should be reviewed:

• the numbers of Premium seat candidates and Premium seats that do not justify their cost
• potential monthly savings from reviewing seats
• a seat table showing price, consumption for the period and per month, active days, sessions, limit-related rejections, utilization, and recommendation

Recommendations are based on consumption extrapolated to a month. A seat with no activity is recommended for revocation; a Premium seat with consumption below the Standard seat price is recommended for downgrade. A Standard seat becomes a Premium candidate if it has limit-related rejections or monthly consumption above 150 USD. Selecting a row opens the Member Seat: Details panel with a breakdown of the calculation.

6. Subscription Limits

Shows subscription-limit utilization for seats with snapshots from the limit collector:

• the numbers of seats with limit collection, seats in the red zone in the latest snapshot, and seats that reached a limit during the period
• average weekly-window utilization and the average five-hour-window peak
• weekly- and five-hour-window utilization by day: maximum and average across members
• seats by weekly utilization level and five-hour-window peaks by member

The red zone starts at 80% utilization of either the weekly or the five-hour window.

7. Member Limits and Recommendations

Shows member limits and seat recommendations:

• peak utilization of a model-specific limit by member and its comparison with the weekly window
• a limit-utilization table with five-hour and weekly peaks, 429 rejections, and recommendations
• estimated peak load from telemetry for seats without a limit collector: peak five-hour spending compared with the USD equivalent of the limit in the claude_code_seat_pricing lookup

8. Reconciliation with a Spend-Report Equivalent

Shows a comparison of costs estimated from telemetry with an equivalent of a subscription spend report:

• estimated cost, list-price spending (gross), discounted amount due (net), and savings from the discount
• requests, prompt and completion tokens, and average discount
• daily list-price and payable spending, and spending by model
• daily comparison of the estimate and spend-report equivalent, including the difference
• spend-report-equivalent rows: day, member, product, model, requests, tokens, gross, and net

Panels labeled Official read from the claude_code_spend_report lookup. A module job populates this lookup by calculating values from telemetry and list prices. It is not an Anthropic export. The equivalent uses completed UTC days, so metrics marked D-1 lag by one day. The contract discount is set in the ai_observability_lookup_spend_report job.