Data downloads
Pulse publishes the published indices, the per-provider observations behind them, the marketplace and stock-state data, and the raw collection logs. Free under CC-BY 4.0. Files regenerate daily.
The full dataset
Complete bulk export
A single JSON document containing everything Pulse collects: per-provider price assessments, raw collection logs, marketplace supply observations, stock observations, capacity forecasts, plus the provider and GPU-model registries. Refreshed daily.
Top-level keys: assessments, raw_collections, collection_runs, providers, gpu_models, instance_gpu_map, supply_observations, stock_observations, capacity_forecasts, date_range.
Per-index series
Convenience slices: one CSV and one JSON per published index. Same content as /api/indices/{slug}.
| Index | CSV | JSON | Last regenerated |
|---|---|---|---|
| Pulse Inference Token Index | CSV | JSON | — |
| Pulse H100 SXM Hyperscaler OD | CSV | JSON | — |
| Pulse H100 SXM Neocloud OD | CSV | JSON | — |
| Pulse A100 80GB Hyperscaler OD | CSV | JSON | — |
| Pulse A100 80GB Neocloud OD | CSV | JSON | — |
| Pulse B200 Neocloud OD | CSV | JSON | — |
| Pulse H200 141GB Neocloud OD | CSV | JSON | — |
| Pulse H100 PCIe Neocloud OD | CSV | JSON | — |
| Pulse A100 80GB Neocloud Spot | CSV | JSON | — |
License
Pulse-owned outputs (index values, historical series, and redistributable sample data) are released under CC-BY 4.0. Provider-origin raw snapshots are not implicitly covered — see what is open and free.
File schema
Every CSV and JSON download follows the same schema. Files are regenerated daily; the timestamp at the top of each file marks the regeneration time.
Columns
| Column | Type | Unit | Notes |
|---|---|---|---|
assessed_at | ISO-8601 datetime, UTC | — | Canonical assessment time. Anchored to 18:00 UTC for daily series; the exact convention per index is documented on the index methodology page. |
value | number | USD per GPU-hour, or USD per million tokens (per index) | The published headline value for the index on this assessment date. |
methodology_version | string | — | The methodology version that produced this value (e.g. v1.0). Permanently citable at /methodology/{slug}/{version}/. |
provider_count | integer | — | Number of providers contributing to this assessment after eligibility filtering. |
p25 / p75 | number | same as value | Distribution shape. Optional; absent for series whose distribution is not published. |
is_carried_forward | boolean | — | true if any contributing provider's assessment was carried forward from a recent prior collection (within the 3-day staleness window). |
note | string, nullable | — | Free-text annotation when a value carries a caveat (e.g. "thin data", restatement reference). |
Conventions
- Timestamp: ISO-8601 with explicit
Z(UTC). Never local time. - Unit: stated in the column header of the CSV (and as the
unittop-level field in the JSON). Currency is USD throughout. - Missing values: empty cell in CSV; explicit
nullin JSON. No interpolation, no zero-fill. A series can simply not publish on a day when publishability thresholds are not met. - Carry-forward: at the index calculation layer only. The underlying raw collections never contain synthetic rows. The
is_carried_forwardflag tells you whether any contributing provider relied on a prior assessment within the documented staleness window. - Regeneration: files are regenerated once daily after the assessment cycle completes. Last regeneration time is the
last_modifiedHTTP header on the file and thegenerated_atfield in the JSON.
API
For programmatic access, see the API documentation.