AI Explorer
One lens for LLM usage across every provider — by model, project, and user, measured in tokens and API-equivalent value.
What is the AI Explorer?
The AI Explorer is a cross-provider view of your LLM usage. Where the Data Explorer answers "what did we spend?", the AI Explorer answers "what did we use?" — every model, from every connected provider and tool, normalised into one view and measured two ways: tokens and API-equivalent usage value.
Model names are standardised across providers, so gpt-5 used via the OpenAI API and the same model used through a coding tool roll up to one row. Usage appears once a token-carrying provider (for example Claude Code, Cursor, or OpenAI) has synced.
Total usage value
$8,214 ≈ API
↑ 12% vs prior period
Total tokens
1.9B
Last 30 days
Top driver
claude-sonnet-5
41% of usage value
| Model | Provider | Tokens | Usage value |
|---|---|---|---|
| ▾ claude-sonnet-5 | Claude Code | 812M | $3,370 |
| Input 61% · Output 9% · Cache read 27% · Cache write 3% — blended ≈ $4.15 / 1M tokens | |||
| gpt-5 | OpenAI | 414M | $2,905 |
| gemini-2.5-pro | GCP / Vertex | 512M | $1,420 |
| gpt-5-mini | OpenAI | 162M | $519 |
Dimensions and measures
| Control | Options |
|---|---|
| Group by | Provider, Model, Project, or User |
| Measure | Usage value ($) or Tokens |
| Chart | Stacked bar, line, or area |
| Granularity | Daily, weekly, or monthly (auto-selected by window length) |
| Filters | Provider, model, project, and user multi-selects plus date presets or a custom range |
| Export | CSV or JSON of the pivoted table |
Direction drill-down
Models with a captured direction split expand to show input, output, cache read, and cache write token counts with their percentage shares, billed dollars per direction where the vendor bills that way, and a blended effective rate (≈ $ per 1M tokens) at your actual mix.
The mix matters because output tokens typically cost several times more than input tokens, and cache reads cost a fraction of fresh input. Two teams using the same model with different mixes can have very different effective rates — which is also why Model Recommendations price alternatives at your mix rather than at headline rates.
Provider coverage
Coverage varies by provider, and the explorer is explicit about it — per-provider flags show whether model, user, project, and priced data are available rather than papering over gaps.
- Full token telemetry: OpenAI, Anthropic/Claude (API and Claude Code), Cursor, GCP Vertex/Gemini.
- Value only: Grok (xAI) reports spend without token counts, so it appears in the value measure but not the token measure.
- Excluded from the model lens: GPU/compute billing (for example Hugging Face Endpoints) and non-generative models (embeddings, speech, image) are kept out of token totals.
Usage with no available list price shows tokens with no dollar value — never a fake $0 — and the affected models are listed in a footer note.