Top 6 tools for tracking AI and token spend

1Vantage

Vantage (private). vantage.sh

Cost management platform that pulls cloud, SaaS and AI provider bills into one set of reports, budgets and anomaly alerts.

Best for

FinOps teams; the site has separate pages for developers, finance teams, management and MSPs.

What it does

• Anthropic integration reports Claude API cost by model, API key and workspace (Admin API key), and Claude.ai usage including Claude Chat and Claude Code by model and user email (Enterprise Analytics API key).
• OpenAI integration reports Platform cost by operation, model, project and API key (tagged with key owner where available), and OpenAI Codex cost by client, model and token type with the user's email as the resource, so Codex spend can be reported per person.

AI features

LLM Token Allocation / Token Cost Allocation Specification

Provider-neutral JSON record per LLM request (provider, model, token usage, optional allocation tags) that Vantage joins to provider bills to split cost by team, customer or feature. Supported at launch for OpenAI, Anthropic, AWS Bedrock, Google Cloud (Vertex AI Gemini and Marketplace Claude) and Azure OpenAI (direct Azure integrations only). Enabled on request through Sales or support.

Connects to

AWS, Azure, Google Cloud, Oracle Cloud, Cloudflare, OpenAI, Anthropic, Cursor, Anyscale, Baseten, Fireworks AI, Modal, ElevenLabs, Cartesia, Datadog, Snowflake, Databricks, Kubernetes, GitHub.

Pricing

Starter free (up to $2,500 tracked cloud spend per month, 3 users, 6 months retention).

Watch out for

The Anthropic integration does not cover Claude hosted on other clouds such as Amazon Bedrock, does not support Claude.ai Pro or Team plans, and only has Claude.ai data from 1 January 2026.

2CloudZero

CloudZero (private). cloudzero.com

Cloud and AI cost platform that allocates spend to teams, products and units of business output.

Best for

The people who share responsibility for AI and cloud spend: finance leaders (CFOs, FP&A, controllers), engineering leaders (CTOs, VPs of Engineering) and platform or cloud ops teams.

What it does

• OpenAI connection (read-only Admin API key) tracks costs by project, model, service tier and usage type; Anthropic Platform tracks cost by project, model and operation for any Claude Console organization.
• Anthropic Enterprise tracks per-user cost by product, model and context window, and Cursor tracks per-user usage, billing groups and model costs. Both require an Enterprise plan.

AI features

AI Hub

Shows what each model call, agent and workload costs.

Connects to

OpenAI, Anthropic, Anthropic Enterprise, Cursor, Amazon Bedrock (via AWS CUR 2.0).

Pricing

No list prices published.

Watch out for

Bedrock spend by IAM principal is shown as private preview in CloudZero's changelog (contact CloudZero to enable), even though a June 2026 blog post describes it as shipped.

3Finout

Finout (private). finout.io

FinOps platform that combines cloud, Kubernetes, SaaS and AI provider bills in one billing view (MegaBill) and allocates them with virtual tags.

Best for

FinOps teams that report AI and cloud spend to finance, leadership and engineering.

What it does

• OpenAI Platform data by model, token type, user ID, project, API key and organization ID; OpenAI Codex data by model, user email, speed and actor type.
• Anthropic connects with an Admin API key, with daily refresh and historical backfill; spend breaks down by model, token type and feature, and Enterprise customers can add the Analytics API for per-user Claude chat, Claude Code, Cowork and Office Agents usage.

AI features

Billy

AI FinOps assistant.

Connects to

AWS, GCP, Azure, OCI, Kubernetes, OpenAI, OpenAI Codex, Anthropic, Bedrock, Vertex AI, Perplexity, fal.ai, Cursor, GitHub Copilot, CodeRabbit, Snowflake, Databricks, Datadog.

Pricing

No list prices published.

Watch out for

For OpenAI, detailed usage data currently covers Completions and Images only (Embeddings, Vector Stores and other services are planned).

4Datadog Cloud Cost Management and LLM Observability

Datadog, Inc. (NASDAQ: DDOG). docs.datadoghq.com

Two Datadog products that cover AI spend from different ends.

Best for

Companies already running Datadog for monitoring that want AI provider bills and per-request LLM traces in the same tool.

What it does

• AI cost coverage for Amazon Bedrock, Amazon SageMaker, Anthropic, Azure Foundry, Google Gemini, OpenAI, Vertex AI, GitHub Copilot, Cursor and Databricks Mosaic AI, with normalized tags for provider, model, token direction, token category and project.
• Out-of-the-box allocation tags by user (user_id, user_email) and API key for Anthropic and OpenAI, user-level allocation for Cursor, and Tag Pipelines to map user email to teams or business units.

Connects to

OpenAI, Anthropic, Amazon Bedrock, Amazon SageMaker, Azure Foundry, Google Gemini, Vertex AI, GitHub Copilot, Cursor, Databricks Mosaic AI.

Pricing

LLM Observability (Agent Observability): Free tier up to 40K LLM spans per month with 15-day trace retention.

Watch out for

LLM Observability costs are estimates from public prices, not invoice figures; high-cardinality tags such as user IDs are not fully supported on cost metrics, and missing cache token breakdowns cause the standard input rate to be applied to all input tokens.

5Langfuse

Langfuse, part of ClickHouse, Inc.. langfuse.com

Open-source (MIT) LLM engineering platform for tracing, evaluation and prompt management that records token usage and cost for every LLM call.

Best for

Engineering teams building LLM applications and agents, particularly developers who value open source and control over their data.

What it does

• Tracks usage and cost of every LLM call, broken down by usage type (input, output, cached tokens).
• Dashboards show cost by model and across users or tags; the Metrics API can filter cost by user or tags.

Connects to

OpenAI, Anthropic, Amazon Bedrock, Google Vertex AI, Google Gemini, Mistral, Cohere, LiteLLM Proxy, OpenRouter, Portkey, Vercel AI Gateway, Kong API Gateway, Azure API Management.

Pricing

Cloud: Hobby free (50k units per month).

Watch out for

Cost inference by tokenizing input and output is not supported for reasoning models, so their cost only appears if token counts are ingested from the provider response; cost is only inferred when the model name matches a model definition, and overlapping usage buckets get double-counted.

6Portkey (Prisma AIRS AI Gateway)

Palo Alto Networks (acquired Portkey in 2026). portkey.ai

AI gateway that routes application calls to LLM providers through one API and adds observability, guardrails, governance and prompt management.

Best for

AI and platform teams running generative AI in production.

What it does

• Cost budgets, token limits and rate limits per API key, per workspace and per provider integration, plus usage limit policies grouped by user, customer, model or provider metadata.
• Requests that hit a budget, token or request cap are blocked before they reach the provider (HTTP 412); limits reset weekly, monthly, every N days or never.

Connects to

OpenAI, Anthropic, Azure OpenAI, AWS Bedrock, Vertex AI.

Pricing

Open Source self-hosted gateway.

Watch out for

On the free Developer plan, logs beyond 10k per month are not recorded and logs are kept 3 days, so cost history is thin; the pricing page lists detailed budget and rate limits under Enterprise.