Picking the right AI model used to mean reading half a dozen pricing pages, three benchmark leaderboards, and a Twitter thread about whether the latest release is actually better. AIpricly collapses that into one screen.
Three things make this site different from the dozens of AI comparison pages already published:
- Decision-first homepage. Three direct answers — Cheapest, Strongest, Best Value — instead of a 300-row table that pushes the decision back onto you.
- Scenario-routed gateway (Phase 2). We don't make you pick a model — we ship named scenarios with primary + fallback chains. The chain is the product.
- Multimodal coverage from day one. Text, image, video, voice, embedding, rerank — most price sites only do text.
Methodology
Pricing comes from OpenRouter as the primary feed, cross-checked against LiteLLM's open dataset; any discrepancy greater than 5% breaks the daily build and surfaces in a drift report. Quality numbers come from Artificial Analysis and LMArena — we never run our own benchmarks. Where a score has not yet been verified against a partner snapshot, a small "editorial estimate" badge appears next to it; that's our promise that we won't silently present hand-typed numbers as measurements. The full data-flow walkthrough lives at /methodology.
Two phases
Phase 1 (now) — comparison site + scenario library + calculator. Free, no signup, monetized via affiliate links to OpenRouter / Together / Fireworks (full breakdown at /disclosure). Carbon Ads in low-traffic side regions; Buy Me a Coffee for individual contributors. No paid placements, no sponsored rankings.
Phase 2 — OpenAI-compatible API gateway. One key, scenario-routed primary + fallback chains, transparent token pass-through plus a deposit fee. No model markups. The site will keep the same data view; the API is opt-in.
Anti-features
Some things we will never do: paid model rankings, sponsored placement, self-run benchmarks dressed up as our own work, login walls in Phase 1, web crawlers scraping vendor pricing pages (we use upstreams instead).
Get involved
Spot a wrong number, a missing model, or a broken page? Open an issue on GitHub. Have a scenario that needs covering, or want to chip in on quality-scoring methodology? Same place. For everything else, the contact is our newsletter inbox.