How Price Gouge builds estimates
How recommendations are built from official provider prices, workload assumptions, deterministic cost math, and optional LLM help.
Audience
Developers and buyers
Metric
Engine rules
Sources
4 official links
What sets the price
Price Gouge starts with official provider prices and the workload fields a user supplies. The CostComparison model turns those inputs into weekly, monthly, and yearly strategy costs for open-model APIs, batch, subscriptions, hybrid routers, AI gateways, managed cloud AI, rented GPUs, local inference, fine-tuning, and self-hosting.
The engine prices the same workload several ways, then ranks the results against the budget and quality assumptions. The recommendation is a route, not a single model name.
- official provider prices set token, GPU, subscription, storage, egress, and credit costs
- structured fields such as budget, documents, tokens, cache rate, GPU hours, storage, and egress override inferred assumptions
- CostComparison owns every numeric strategy cost shown in the UI
- pricingNote reformats the chosen strategy into short sections that are easier to read
Where the LLM fits
The LLM can classify the workload when a user writes a short request such as "process one million documents" or "keep this under $500/month." It can also write a compact explanation after the deterministic comparison exists.
The LLM cannot set prices, create provider records, choose billing units, or change strategy cost components. If the LLM fails or returns weak parsing, rules still produce a complete estimate from the available fields.
- Allowed: workload classification, missing-input hints, and short advisor prose.
- Blocked: numeric prices, provider source records, billing units, cost components, and official source dates.
- Rendered UI should show structured costs before generated explanation.
Recommended route logic
The default policy is hybrid routing. Routine inference goes to cheap open-model, batch, local, or GPU-hosted lanes. Frontier systems such as Claude, Codex, OpenAI, and Gemini stay available for planning, architecture, difficult bugs, research synthesis, code review, and high-risk reasoning.
The cheapest lane is useful only after quality clears the bar. A route that causes retries, long outputs, low accepted-output rate, or missed policy checks can cost more than the table price suggests.
- Use open-model APIs first for extraction, classification, summaries, drafts, and repeated agent loops.
- Use batch routes for backfills, document jobs, evals, and offline enrichment.
- Use frontier routes for judgment calls where the answer changes a high-value decision.
- Price GPU hosting after utilization, storage, egress, reliability, and operations are measured.
API path
Use /api/public/engine to inspect the engine contract, supported pricing models, public endpoints, source freshness, and LLM permissions. Use /api/public/pricing when a user needs a priced answer for a workload.
curl https://pricegouge.me/api/public/engine
curl -X POST https://pricegouge.me/api/public/pricing \
-H 'Content-Type: application/json' \
-d '{"q":"What is the cheapest production architecture under $500/month?"}'