Introducing Vantage Tokenomics: Measure and Optimize Token Spend

Measure and optimize token spend, track how efficiently each person uses tokens, and see the value of completed work like GitHub pull requests and Linear issues, with AI usage attributed to identity, purpose, and outcome in real time.

Introducing Vantage Tokenomics: Measure and Optimize Token Spend
Author:Vantage Team
Vantage Team

Today we are introducing Vantage Tokenomics: a new set of tooling to measure and optimize token spend, track the efficiency of each person’s token usage, and show the value of completed tasks, like GitHub Pull Requests or Linear Issues, all represented in Vantage in real time. “Tokenomics” is now a new dedicated portion of the Vantage console and supports all major AI IDEs (Cursor, Claude Code, Codex, etc.) and all major AI Gateways (OpenRouter, LiteLLM, etc.).

It’s almost too obvious to say that AI has become ubiquitous: every organization is aggressively leveraging AI. To date, Vantage has mainly been focused on helping organizations manage the token costs associated with running workloads, or things that predominantly impact COGS or have an impact on gross margin. While Vantage could show token spend in aggregate by each developer on the team, the context behind what drives those costs, and how efficiently it was accomplished, were still opaque. With AI costs being under the most critical lens for nearly every organization, it is more important than ever to understand the value delivered and outcome associated with usage.

Now, with Vantage Tokenomics, customers can automatically attribute token costs to identity, purpose, and outcome, allowing them to understand what the spend drivers are and what value is delivered. Vantage Tokenomics works by ingesting OpenTelemetry logs, so any major AI IDE or Gateway is securely supported out of the box, with no applications installed on developer laptops. Once logs begin syncing, three new portions of the console are available:

  • Roster: A roster of all identities generating token spend, where you can view every identity’s current token spend and sort by estimated gross spend or by the number of GitHub PRs and Linear Issues completed. Selecting an individual’s name gives a breakdown of all the tasks (Linear Issues, GitHub PRs, etc.) they have completed, with respective costs. Anyone who isn’t on the list yet can be invited from the Roster to set up the OpenTelemetry stream and the Classifier.
  • Classifier: Vantage Tokenomics can automatically tag prompts as they happen so you can tie them to things like conversation purpose, GitHub Pull Requests, Linear Issues, or any other arbitrary metadata. You can customize your organization’s tagging strategy on the Classifier page in the Vantage console.
  • Tokenomics Report: Customizable reports provide a deeper view of token usage, letting teams combine token consumption and estimated gross costs with classifier tags to explore the workflows and patterns driving spend. Efficiency metrics such as input/output ratio, cache hit rate, and cost per million tokens help teams investigate spikes surfaced in the Roster and identify opportunities to optimize.

Tokenomics is being announced in private preview for select Vantage design partners. To request access, please reach out to support@vantage.sh. For organizations that are not currently Vantage customers that are interested in utilizing Tokenomics, reach out to sales@vantage.sh or request a demo here.

Frequently Asked Questions

1. What is being launched today?

Vantage is launching Vantage Tokenomics, a new set of features for measuring, optimizing, and justifying AI spend. Vantage Tokenomics has three components: the Roster, the Classifier, and the Tokenomics Report. Together, they attribute estimated AI cost and usage, automatically tag the identity, purpose, and outcome of that spend, and show how efficiently it is being used, so customers can see where their R&D spend is going and what value it drives.

2. Who has access to Vantage Tokenomics?

Vantage Tokenomics is in Private Preview for select Vantage design partners and will open in Beta to existing customers soon. To be added to the Beta, reach out to support@vantage.sh or your Vantage account team. Organizations that are not yet Vantage customers and want to use Vantage Tokenomics can reach out to sales@vantage.sh.

3. How much does this cost?

There is no cost for Vantage Tokenomics while it is in Private Preview. Usage-based pricing will be introduced at a later date. Using the Classifier will cause an increase in customers’ AI provider token usage.

4. What is the Roster?

The Roster is the homepage of the Tokenomics experience, and acts as a list of every user in your organization generating token spend sorted by estimated gross spend, and also shows attributes such as tokens consumed and outcome-based attributes like the number of GitHub pull requests and Linear issues completed for the trailing 30 days. Clicking a name shows which tasks that identity has completed, along with the cost of each. Anyone who isn't on the list can be invited from the Roster to set up the OpenTelemetry stream and the Classifier.

5. What is the Classifier?

The Classifier applies your organization's tagging strategy to each request automatically, including when data arrives through a gateway. Hooks in your coding tools or gateway fetch your organization's customized tagging skill from Vantage and apply it as work happens, so developers never need to tag by hand. Classifier tags supplement tags already on your data and do not override them by default.

Out of the box, the Classifier applies the following tags:

  • Identity: the user the spend belongs to
  • Purpose: what the tokens are being used for, such as development, research, design, or bug fixes
  • Outcome-based tags applicable to your usage: the result of the spend
    • Pull request ID
    • Branch
    • Repository
    • Ticket ID from ticketing systems like Jira or Linear to roll up cost to a project or feature

These are applied to any usage and can have their guidance edited, but cannot be deleted. Up to 10 additional tag keys can be added into Tokenomics Reports, such as Project, Business Unit, or Team.

Because tagging happens per request, a session that mixes different kinds of work is split across the identity, purpose, and outcome that apply to each request.

6. What is a Tokenomics Report?

The Tokenomics Report is a detailed visual representation of your organization’s AI consumption that allows organizations to investigate, optimize, and tie AI usage to outcomes. The report shows the estimated cost, usage, and efficiency, attributed back to the user, purpose, and outcomes that drove it, such as GitHub pull requests and Linear issues. Efficiency is measured with cost per million tokens (Avg. Cost / 1M), input/output ratio (Input : Output Ratio), and cache hit rate (Avg. Cache Hit Rate).

Select a user to open their developer page, which shows that developer's consolidated usage and efficiency metrics along with their unique sessions and unique outcomes.

7. I’m just beginning to use Vantage Tokenomics, how do I get started?

Users with Vantage Owner or Integration Owner permissions will navigate to the Tokenomics section of Vantage, and have the ability to flip on Tokenomics for their account. After that you will be prompted to connect your coding tools, organization-wide gateway, or a mix of both, depending on your coverage. Vantage will provide onboarding instructions for each of your integration sources.

Once logs begin to populate, the Roster will be your default page. From there, you can navigate to Tokenomics Reporting and create custom views of your log data to see who is spending, what they’re spending on, and the outcomes.

8. How do I set up Tokenomics if my Organization uses an AI Gateway?

When adding a gateway, either through Tokenomics Onboarding or through the Vantage Integrations page, you will be prompted to provision an OpenTelemetry Ingest Token, and add that to the configuration of each Gateway (see documentation of supported gateways for instructions on each). An administrator can complete this setup for the whole organization, with no software installed on developer laptops.

Once completed, you will be guided to configure the hook to import your tagging skill. This can also be performed one time per gateway. Skill URLs are static, so any time you update your tagging skill, no config changes are needed.

9. How do I set up Tokenomics if my Organization does not use an AI Gateway?

You will set up Tokenomics using a prompt and our plugin made available in our finops-as-code GitHub marketplace.

If you have an Enterprise subscription, for each provider you can follow the below setup instructions.

  • Anthropic/Claude: an enterprise admin will run:
claude plugin install vantage-tokenomics --marketplace https://github.com/vantage-sh/finops-as-code/tokenomics/anthropic/plugin
  • OpenAI/Codex: an enterprise admin will run:
codex plugin marketplace add https://github.com/vantage-sh/finops-as-code.git \
--sparse tokenomics/codex/plugin
codex plugin add vantage-tokenomics --marketplace vantage
  • Cursor: an enterprise admin will navigate in the admin page to Dashboard → Plugins & MCPs → Add Marketplace → Import from Repo, and install vantage-tokenomics and mark it Required for the team.

If you have a non-Enterprise subscription (e.g., Teams or Individual) for Anthropic, Cursor, or OpenAI, installation is offered in one of two ways:

  1. Vantage MCP users can call setup_vantage_tokenomics and the MCP will guide their agent through plugin installation and configuration. At any time they will also be able to run check_vantage_tokenomics to get insight into setup and configuration.

or

  1. Share a prompt with the user to set up Tokenomics:

    Install Vantage Tokenomics for this coding agent using the setup instructions at <GitHub URL to SETUP.md>. Follow the instructions to install the plugin at user scope, configure metadata-only usage export, and start browser authentication if needed. Never print credentials or include them in chat. Verify the connection and tell me if a restart or hook approval is required.

10. What tools and gateways are supported?

Any AI tool that supports OpenTelemetry (OTel) is compatible with Vantage Tokenomics, so support grows as the industry adopts open standards. Most common tools support OpenTelemetry today:

Coding tools:

  • Cursor
  • Claude Code
  • Codex

AI gateways:

  • OpenRouter
  • LiteLLM
  • Bifrost

If you use a tool that isn’t currently supported, submit a feature request to Vantage through your account team or by emailing support@vantage.sh.

11. Where do I manage the Classifier, and how do I make changes?

Administrators define the organization's tagging strategy on the Classifier page in the Vantage console, or programmatically with Terraform, the API, or the Vantage MCP. The Classifier provides a default prompt that creates the following set of tags: identity, purpose, and outcome-based tags (pull request, branch, repository, and ticket ID). Administrators can add up to 10 additional tag keys they want collected on their data, specify allowed attributes, and provide context around how to determine tag values. This is all defined through natural language, and the Vantage Agent translates it to a skill and provides it as a unique URL for your AI gateway or AI IDEs to point to. When you update your tagging strategy, the updated version is used automatically when new requests are generated. Changes apply to future requests and do not reprocess previously tagged costs.

12. Does Vantage install anything on developer laptops?

Vantage installs hooks, skills, and configuration that enable tools such as Claude Code, Codex, and Cursor to send OpenTelemetry data to Vantage for reporting. It does not install applications or script code to run locally on developer laptops, and it does not intercept network traffic. Provider admin APIs supplement reporting when a tool’s native export is unavailable.

13. What can I group and filter by?

In the Tokenomics Report, you can group and filter by model, provider, source, token type, and any tags such as the managed tags (identity, purpose, and outcome tags) or the 10 tags your organization has added, such as project, team, or cost center.

14. How are efficiency metrics calculated?

  • Cost per 1M tokens: (total estimated token cost ÷ total tokens) × 1,000,000. A blended rate across ordinary input, cache reads, cache writes, and output, priced at the applicable public or customer-supplied model rates. Total tokens equals total input plus output, counting each token once.
  • Cache hit rate: (cache-read input tokens ÷ total input tokens) × 100. The percentage of input served from cache. Total input includes ordinary input, cache reads, and cache writes; cache writes are not cache hits.
  • Input/output ratio: total input tokens ÷ output tokens, displayed as input tokens per output token (for example, 100:1). Includes cache reads and writes. This describes context consumption relative to generation; a high ratio alone does not indicate waste, especially when context is served from cache.

15. How are estimated costs calculated?

If a provider or gateway supplies estimated costs, Vantage will use that as the Gross Cost for the usage. For vendors that do not provide a cost, estimated gross spend applies public model rates to reported token usage, by token type where available. "Gross" means list price before discounts, and "estimated" means it has not been reconciled with the provider's bill. It does not represent your final net bill after discounts, credits, or subscription allowances.

16. How fresh is the data, and how long is it retained?

AI usage appears in Vantage within minutes of the usage occurring. Vantage retains Tokenomics data at a 15-minute granularity for 7 days. After that, the data is compacted into daily views, which are retained for your Vantage account's retention period.

17. Can I backfill historical data?

No. Vantage Tokenomics begins collecting data once a source is connected and sending OTel logs to Vantage. It cannot backfill usage from before the source was connected.

18. Does Vantage collect prompts and completions?

No. Vantage Tokenomics uses usage and attribution metadata and does not store prompt or completion text. We recommend disabling content export in your gateway or developer tool. If content is sent anyway, Vantage may receive it briefly during processing and strips it before storage. This data is never used to train models.

19. How does this relate to Cost Reports and LLM Enrichment?

Cost Reports and Vantage Tokenomics serve different purposes, but each helps tell the AI spending story. LLM Enrichment enables Finance, FinOps, and Engineering teams to allocate provider-billed costs, supporting budgets, chargeback or showback, and spend forecasting. Tokenomics provides real-time visibility into usage across the same AI providers and helps teams address inefficiencies as they happen.

Cost Reports are built for allocation, showback, and chargeback, using billed cost data that comes directly from your providers. With LLM Enrichment, Cost Reports can also use the same request logs that power the Tokenomics Report to allocate those billed costs.

Vantage Tokenomics is built to tie usage back to outcomes and optimize it as it happens in real time, so teams can see what their AI spend produced and act on inefficiencies within minutes.

Sign up for a free trial.

Get started with tracking your cloud costs.

Sign up

TakeCtrlof YourCloud Costs

You've probably burned $0 just thinking about it. Time to act.

Introducing Vantage Tokenomics: Measure and Optimize Token Spend | Vantage