Getting more value from Claude Code

Anthropic reports a great deal about how Claude Code is used: sessions, accepted lines, pull requests, tokens and cost, for each person and model. It also gives admins limits, seat types and model settings to act on. What your team adds is the business view, meaning which team each dollar belongs to and what the work produced in your code host.

We suggest you give each dollar an owner first, then choose the settings that fit each team, and then check each change against merged pull requests and the bill. Prices change, so the ones here are from Anthropic's pages in October 2026.

What Anthropic already gives admins

Anthropic's analytics guide describes what each plan reports.

• On Team and Enterprise plans, a dashboard shows daily active users, sessions, accepted lines and how often people accept a suggestion. With the Claude GitHub app installed, it also counts merged pull requests that contain code written with Claude Code, and labels them claude-code-assisted in GitHub. Anthropic calls these counts "deliberately conservative".
• On the Claude Console, a dashboard shows spend and accepted lines for each member. The Claude Code Analytics API returns the same data per person and day: sessions, lines added and removed, commits, pull requests, and tokens and estimated cost by model. It is free with an Admin API key.
• Controls. On Team and Enterprise, each seat has an included allowance. Past it, usage credits apply, with spend limits for the organization, a group or one member. Enterprise admins also get alerts at 75% and 90% of the organization's limit (Anthropic, July 2, 2026), a default model, and a maximum effort level for each role.
• Seat types. On the Team plan, a Standard seat is $25 a month, or $20 billed annually. A Premium seat is $125, or $100, with five times the usage. Enterprise is $20 a seat a month billed annually, plus usage at API rates.

Three ways to pay, and where each one reports

Anthropic's cost guide says each developer is metered by the way they signed in. A Team or Enterprise seat draws on its allowance and reports in the claude.ai admin console. A Claude Console sign-in is billed per token and reports in the Console. Claude Code on Amazon Bedrock, Google Cloud or Microsoft Foundry is billed by that cloud, and Anthropic's analytics do not cover it. If your company uses more than one, add them together before anyone quotes a total.

What the reports leave out

The dollar figures in /usage, the Console dashboard and OpenTelemetry are estimates at list price, and Anthropic points to the billing page for actual costs. An admin can set the modelPricing managed setting to your contracted rates, which brings the estimates closer, though finance will still go by the invoice.

Accepted lines tell you Claude Code is in use, but the analytics guide notes that this count does not follow lines deleted later, so it cannot tell you what shipped. Each report also names a person or an API key, without their team or what one merged change cost.

Valistry is built for that part: it reports Claude Code spend by team, and joins it to merged pull requests from your code host.

Give each dollar an owner

The owner of Claude Code spend is usually a team, and each way of buying gives you a way to find it.

1. 1The Console. Anthropic creates one Claude Code workspace and gives each member a key of their own in it at sign-in. Every dollar arrives with a person, so you map each person to a team once, from your directory. This workspace is also the only one with a monthly spend limit for each person.
2. 2Team and Enterprise. Put people in groups that match your teams. Enterprise admins can bring groups from the identity provider with SCIM, and see usage and cost by group.
3. 3Other Claude API work. Create a workspace for each team, up to 100 by default, and issue keys inside it. The usage and cost reports group by workspace.
4. 4Cloud providers. Send OpenTelemetry from each developer's machine, or use a gateway, and add the team there.

We suggest reporting by team by default. A person's own figures help that person work better, and a team view answers the budget question without ranking colleagues.

The levers Anthropic offers, and what each saves

Model choice. Claude Code uses Opus 5.5 by default on Anthropic plans and the API, says the model configuration guide. The cost guide suggests matching the model to the job: Sonnet for most coding, and Opus for complex design decisions and reasoning over many steps. On the pricing page, Opus 5.5 lists at $4 per million input tokens and $20 per million output tokens, and Sonnet 5.5 at $2 and $10. A cache read costs $0.20 per million on both, so the two differ in new input and in output.

• Admins can set the default model, or the models people may choose, in managed settings. The opusplan setting uses Opus while planning and Sonnet while writing the code.
• Haiku 4.5, at $1 and $5, suits simple subagent work at a quarter of Opus 5.5's price.
• Effort. Opus 5.5 and Sonnet 5.5 default to medium effort. Thinking tokens are billed as output, so a lower effort on simple work uses fewer of them. Anthropic publishes no saving figure.

Caching. Claude Code caches the repeated part of each request by itself. A cache read costs 10% of the input price, and 5% on Opus 5.5. Each model has its own cache, so a model switch mid-session reads the whole conversation again without the discount, and it is best to choose the model when the session starts. On an API key, the cache lasts five minutes by default. The promptCacheTtl setting raises that to one hour, and because a one-hour write costs twice the input price, it pays back after two reads.

Context. The cost guide says unexpectedly high spend usually comes from long sessions that were never cleared, or from Opus left as the default model. Clearing the session between unrelated tasks and turning off unused MCP servers both help, though Anthropic publishes no figure for either.

Seats. A Premium seat costs five times a Standard seat and gives five times the usage, so it fits people who often reach the Standard allowance. Anthropic publishes no figure for unused seats, and a seat nobody used last month is worth a conversation with its owner.

What costs more. Fast mode, turned on with /fast, makes Opus up to 2.5 times faster, at $8 and $40 per million on Opus 5.5, twice the standard price. It is off on Team and Enterprise until an Owner turns it on, and it suits live debugging better than long background work. Agent teams use about seven times the tokens of a normal session when teammates run in plan mode, says the cost guide.

How to know a change worked

A change has worked when the same team ships the same kind of work for less, with no drop in quality. You can measure that in four steps.

1. 1Pick one team and one change, for example Sonnet 5.5 as the default model for the platform team.
2. 2Write down the baseline over 30 days. Take the team's cost from the bill: Console spend for its people, or its seat fees and usage credits. Count its merged pull requests in the code host; in GitHub, a search with `is:pr is:merged` and a merged: date gives the count. Divide the cost by the merged changes.
3. 3Make the change, and leave everything else as it was.
4. 4Measure the same team over the same number of days, and compare the cost per merged change.

Quality is part of the test. Watch reverted changes, failed builds and rounds of review next to the cost, because if those go up, a cheaper model can cost more in review time than it saves on the bill.

What one merged change costs on your bill tells you more than any count of lines.

On a seat plan, a cheaper model may not change the seat fee. It makes each seat's allowance go further instead, so the saving shows up as fewer usage credits or fewer Premium seats, and that is the line to measure.

What Valistry adds

Valistry reads Anthropic's usage and cost reports and the Claude Code Analytics API, and never reads a prompt. It stores each person as a code in place of an email, so you can check cost per merged change for each team every month without ranking people. The Anthropic page lists what it reads.

The teams that get the most from Claude Code usually know who spent each dollar, which settings fit their work, and what one merged change costs, and the data for all three is already in Anthropic's reports and your code host.