Product guide · market index
Use the index to separate the current published GPU cloud price level from historical movement. Read provider-balanced medians, provider dispersion, raw ranges and the fixed matched-cohort line as different signals. The index is a market-observation tool, not a transaction-weighted price, capacity guarantee or forecast.
Reading order
The current model table and historical index answer different questions. Start with the current provider-balanced market level, then inspect dispersion, and only then interpret the fixed-cohort history.
1 · Coverage
Start with eligible offers, providers, model labels and matched-cohort coverage. A larger current catalog improves current market breadth, but new catalog rows do not automatically enter historical points.
2 · Current level
The current summary first collapses eligible rows within each provider and then gives providers equal influence. This prevents a provider with many regions or SKUs from dominating the summary merely because it publishes more rows.
3 · Provider dispersion
The provider P25–P75 range describes the middle part of provider-level price observations. Use it to understand dispersion around the balanced median rather than treating the median as if every provider quoted the same rate.
4 · Raw range
The model table also preserves the raw observed minimum and maximum. Those endpoints can reflect region, SKU, node-size or purchase-context differences, so they are evidence boundaries rather than automatic recommendations.
5 · Historical line
The historical index uses the same matched offer identities across its retained dates. Baseline 100 is a reference level: movement above or below it describes the price direction of that fixed cohort, not the entire changing catalog.
6 · Drilldown
Use model pages, price history, alerts and the global comparator when the decision depends on a specific GPU, provider, region, purchase model or availability signal.
Metric reference
Keep current cross-sectional summaries separate from historical matched-cohort movement when citing the index.
| Signal | What it describes | Do not infer |
|---|---|---|
| Provider-balanced median | A current price level after reducing provider row-count bias and giving providers equal influence. | A transaction-weighted market price, invoice average or capacity-weighted rate. |
| Provider P25–P75 | The middle spread of provider-level current price observations. | A confidence interval, forecast range or guaranteed purchasable band. |
| Raw range | The lowest and highest eligible normalized rows in the current model slice. | Interchangeable configurations or universal market floor and ceiling. |
| Index level | Fixed-cohort price movement relative to the baseline value of 100. | The current provider-balanced median expressed as a percentage. |
| Matched cohort | The offer identities retained consistently across the historical comparison. | All providers, all current offers or an automatically expanding historical sample. |
| Deployability evidenced | The share of current price rows with a separate qualifying deployment-eligibility signal. | That the same share of capacity is in live stock or currently rentable. |
| 730 GPU-hour figure | The provider-balanced median multiplied by a continuous-month GPU-hour horizon. | A complete monthly invoice including storage, egress, support, taxes or idle overhead. |
Provider balancing
Cloud providers publish very different catalog shapes. One provider can expose many regions, instance variants or duplicated price points while another exposes only a few. Counting every row equally would allow catalog verbosity to become market weight.
Within provider
The index derives provider-level summaries before combining providers. This keeps a large regional catalog from receiving extra votes solely because it contains more rows.
Across providers
The result is useful as an independent published-price signal. It intentionally does not estimate provider market share, customer spend, transaction volume or installed capacity.
Historical contract
A historical line becomes misleading if a cheaper or more expensive newly discovered provider can rewrite old points. SaaS Sentinel therefore separates today's broad catalog from the smaller set of identities that can be compared consistently across retained historical dates.
Current catalog
New qualified providers and offers can improve the current model summaries without pretending those rows existed in earlier observations.
Fixed cohort
The same offer identities support each historical point. A flat line means no distinct normalized rate changed inside that cohort; it does not mean every cloud quote or capacity state stayed unchanged.
Choose the right surface
Move to the narrowest evidence surface that matches the decision you are actually making.
Market direction
Use the provider-balanced current summaries and fixed-cohort historical line when you need a broad published-price market signal.
One GPU over time
Use the append-only model history when you need retained low/median/high snapshots, individual rate moves, evidence refreshes and source-level auditability.
Purchase shortlist
Use the comparator when provider, exact GPU, region, purchase model, availability evidence and your GPU-hour horizon determine the buying decision.
Monitoring and reuse
Use the interface for interpretation and the existing machine-readable surfaces when another system needs the same evidence.
Alerts
Subscribe to the catalog or model-specific Atom feeds when you want verified price changes without treating evidence-only refreshes as movement.
API
The public API exposes the index with generated date, coverage, methodology version, source hashes and limitations so downstream analysis can preserve provenance.
Embed
Use the existing widget when a third-party page needs a lightweight index surface rather than a copied screenshot or manually transcribed value.
Need the raw index?
When citing or reusing the index, keep the generated date and evidence boundary so a historical number is not presented as a timeless market fact.