The reputation index for AI infrastructure Vol. I · No. 1 · 10 July 2026

DatumIndex

Sine ira et studio — the record, without favour

Editorial Standards

Methodology

How we arrive at a reputation, what our figures do and do not mean, and precisely how far each one should be trusted. Read this before quoting anything on this site.

Datum Index publishes a single headline number for each provider — a composite reputation from 0 to 100 — and a great deal of our editorial energy goes into making sure that number is never mistaken for something it is not. It is a judgement, formed editorially from the public record. It is not a measurement we took, not a metric the provider reports, and not a figure to be quoted as fact. This page exists so that nothing on the site reads as fabricated certainty.

The Composite

What the reputation score is

The reputation score is an editorial composite index, not a first-party metric. reputation.score is an editorial composite index (0-100), not a first-party metric. See each site's Methodology page for the weighting model. We weigh, in descending order of influence, four families of signal:

The weighting model

  • Reliability & track record — published uptime, incident history, years in operation, and the accumulated verdict of practitioners. This carries the most weight.
  • Trust & compliance — named attestations (SOC 2, ISO 27001, HIPAA, GDPR posture) and enterprise diligence.
  • Transparency & economics — honest, legible pricing and the absence of billing surprises, read alongside indicative cost.
  • Capability & performance — model quality or breadth, throughput and latency, weighted least because it is the easiest to buy and the fastest to change.

We deliberately do not publish a spurious decimal formula. A composite reputation is a considered editorial reading, and dressing it in false arithmetic would imply a precision we do not claim. The direction of travel (▲ improving, ▼ slipping, — steady) reflects the trend in reviewer sentiment, not a change in the score.

Caveats

On pricing and latency

Two columns on this site are especially easy to misread, so we state their limits plainly.

Pricing. Blended prices are approximate USD per 1M tokens for a representative open model where the provider hosts one; proprietary-only providers list a flagship model instead. Latency figures are order-of-magnitude reference points, not SLAs. Prices are indicative reference points for comparison, not quotes; they move often, and a provider's real bill depends on model, region, context length, caching and commitment. Never cite a price here as a provider's current rate — follow the link to the provider and confirm.

Latency & throughput. Time-to-first-token and tokens-per-second are order-of-magnitude reference points, not SLAs. They vary by model, hardware tier, load and geography, and a single figure cannot capture that distribution. Datum Index does not run first-party probes; we read the public record.

The Confidence Legend

How far to trust each figure

Every provider — and many individual figures — carries one of three confidence marks. This is the legend, reproduced verbatim from our data schema.

high 7 of 20 providers

Figure confirmed against the provider's public pricing/docs page on the last_reviewed date.

medium 12 of 20 providers

Figure based on public sources but not re-confirmed on the exact review date; may drift.

seed 1 of 20 providers

Illustrative seed value pending first-party verification. NOT to be quoted as fact.

Full Disclosure

Our confidence, provider by provider

So that nothing on the site reads as unqualified fact, here is the confidence we place in every dossier and its pricing figure.

Confidence disclosure · reviewed 10 July 2026
Provider Category Dossier confidence Pricing confidence
OpenAI Frontier Lab high medium
Anthropic Frontier Lab high medium
OpenRouter Router high medium
Groq Specialist Hardware high medium
Together AI Aggregator high medium
Fireworks AI Aggregator high medium
Modal GPU Cloud medium seed
Google Vertex AI Hyperscaler medium seed
Baseten GPU Cloud medium seed
Mistral La Plateforme Frontier Lab medium seed
cheapestinference Subscription (flat-rate) seed seed
Cerebras Inference Specialist Hardware medium seed
Perplexity API Specialist Application medium seed
DeepInfra Aggregator high medium
Replicate Aggregator medium seed
SambaNova Cloud Specialist Hardware medium seed
Nebius AI Studio GPU Cloud medium seed
Lepton AI GPU Cloud medium seed
Novita AI Aggregator medium seed
Hyperbolic GPU Cloud medium seed

Where a figure is marked seed, it is illustrative and pending first-party verification — honestly, an estimate that conveys the shape of an offering rather than a confirmed fact. We upgrade these marks in public as verification allows.