| name | cloud-instance-shapes |
| description | Research and compare machine shapes across major clouds and neo-clouds using normalized fields |
Purpose
Compare compute instance shapes across providers with consistent fields, starting from fast aggregators and validating with provider sources.
Primary source
Use this first for broad discovery and filtering.
Secondary sources (validate and fill gaps)
- Official provider docs/pricing pages for AWS, GCP, Azure
- Neo-cloud docs/pricing pages (for example: Runpod, Lambda, CoreWeave, Vast)
- Provider CLI or API docs when shape metadata is unclear
Normalized fields
For each candidate shape, capture:
- provider
- instance_type
- region
- vcpu
- memory_gib
- gpu_count
- gpu_model
- local_nvme_gib
- network_bandwidth_gbps
- on_demand_price_hour_usd
- spot_or_preemptible_price_hour_usd
- source_url
- source_timestamp
If a field is unknown, set N/A and note why.
Workflow
- Clarify constraints (workload type, GPU/CPU needs, min memory, region, budget).
- Pull broad candidates from
instances.vantage.sh.
- Validate top candidates against provider docs for accuracy.
- Return a compact comparison table plus recommendations.
- Highlight tradeoffs (price vs performance vs availability risk).
Output format
- Short recommendation summary (top 2-4 picks)
- Comparison table using normalized fields
- Notes section for caveats and missing data
- Source links for every recommendation
Rules
- Do not infer missing numeric values; mark unknowns explicitly.
- Prefer current provider docs when aggregator and provider values differ.
- Call out region-specific constraints and stock/availability risk.
- Keep output concise and decision-oriented.
