Fetch Foundation Model Pricing

Fetch Foundation Model Pricing

Purpose

Retrieve current public pricing for foundation models across the major AI/cloud providers and return structured, provenance-labeled output. Supports two modes:

• Single-model lookup: fetch the current price for a specific model and deployment target (e.g., Claude Sonnet 4.5 on Anthropic direct, or on Bedrock).
• Comparative table: build a side-by-side price comparison across two or more models or providers for the same task type (text, image, embedding, or GPU-hour).

When to use

Use this skill when:

• The user asks "how much does model X cost per token / per image / per GPU-hour"
• The user wants to compare inference costs across two or more providers for the same model family or equivalent capability tier
• The user needs to estimate monthly AI inference spend given a volume projection
• The user wants to understand how context caching or batch pricing changes the effective cost curve
• The user wants a FOCUS-aware cost breakdown (BilledCost, EffectiveCost) for AI inference line items

Operating rules

• Fetch live prices first. Use WebFetch to retrieve prices from each provider's public pricing page before relying on any internal knowledge. AI model pricing changes frequently; stale numbers mislead.
• Label every price. Each price value must carry exactly one provenance label:
  • live-price — fetched from a provider's public pricing page or API within this session; include source URL and ISO 8601 timestamp.
  • documentation-based — sourced from official documentation when a live fetch was not possible; note the documentation URL and its visible publication date.
  • assumed — derived from an analogous model or tier when no direct published price exists; state the assumption explicitly.
  • excluded — pricing that exists but was intentionally omitted from the output; state why.
• Include source URL and timestamp. For every live-price value, state the exact URL fetched and the UTC timestamp of the fetch to the minute (e.g., 2026-05-13T14:32Z).
• On-demand pricing only unless told otherwise. Do not apply reserved capacity, committed use, or enterprise negotiated pricing unless the user explicitly requests it.
• No credentials required or accepted. All provider pricing pages are public and unauthenticated. Never ask for API keys, billing account IDs, or tenant-specific data.
• FOCUS column mapping. Where a cost estimate is produced, note the corresponding FOCUS v1.2 columns: BilledCost (what the provider charges), EffectiveCost (after credits or discounts), ServiceCategory (AI and Machine Learning), ChargeCategory (Usage), SkuId (model ID + deployment tier), SkuPriceId (price dimension: input-token / output-token / cached-token / image / gpu-hour).
• Load references only when needed.

Pricing dimensions

Providers in scope

Response minimum

Return, at minimum:

• confirmed model name(s), provider(s), and deployment target(s)
• pricing dimensions covered (input, output, cached, batch, image, GPU-hour)
• line-item price table: model | provider | dimension | unit price | provenance label | source URL | fetch timestamp
• for comparative tables: a summary row noting the cheapest and most expensive option per dimension
• key assumptions (on-demand, no reserved/committed pricing, USD unless stated)
• FOCUS column mapping for any cost estimate produced

References

Load these only when needed:

• Provider pricing URLs — public pricing page URLs per provider for live WebFetch.
• Token economics — input/output/cached/batch pricing model, $/M token math, and context caching cost curve.