Skip to content
Sovyron

How Sovyron measures LLM API prices

Sovyron is a read-only price comparison: it republishes list prices from a public catalog, normalizes them to one unit, and ranks them with published rules. This page states the source, the rules and the limits, so any number on the site can be checked.

What Sovyron measures

Metered API prices for 1985 canonical models across 226 providers, in USD per 1M tokens — input, output, cache read and cache write, plus the context window and output limit each provider quotes. Subscription plans are tracked separately and never counted as a per-token price. 632 models are sold by two or more providers and therefore get their own comparison page; 149 of them are widely offered enough to appear on the home page and in the matrix.

Data source

Model names, context limits and per-token prices come from the public models.dev catalog. Every build fetches the catalog, normalizes each provider's model names to one canonical model, then drops providers and duplicates that would double-count the same offer. Flat plan fees cannot be read from that catalog, so they live in the site's own committed table and are matched against vendor pricing pages by a scheduled check; the fees below were last verified 2026-09-13.

What “cheapest” means

A “cheapest” figure is the lowest metered per-token price among the offers eligible for comparison: offers billed by subscription plan, free-tier offers, and offers missing either side of the price are excluded, because a zero or a bundle is not a per-token rate. The remaining offers are ranked independently on input, on output and on cache read, and the cheapest of each is highlighted in the tables. Spread is the highest eligible output price divided by the lowest.

The 3:1 blend

Where one number is needed to compare offers across providers, Sovyron uses a blended rate of 3 input tokens to 1 output token:(3 × input + output) ÷ 4. If a provider quotes only one side, that side is used alone; if both are missing the offer sorts last. The blend is a ranking aid, not a bill — a real invoice depends on the actual input:output mix.

Subscription plans

A plan's effective rate is its monthly fee divided by the tokens a month it is spread over: at 1M tokens a month it equals the fee. Break-even is the monthly token volume at which the plan's fee equals the metered API cost of the same 3:1 blend, so below it the API is cheaper and above it the plan is. Plan quotas are usage windows, not token allowances, so these figures are fee arithmetic over list prices, not vendor quotes; the fee table states the date the prices were verified.

Which models get a page

Every model can appear in a table, but a model gets its own page only when at least two providers list it and at least one of them sells it at a metered per-token price — otherwise there is nothing to compare. The home page and the matrix use a narrower cohort: models offered by at least 12 providers, ranked by how many providers carry them.

How often the numbers change

A scheduled pipeline refreshes the catalog every day and commits the result, so each price carries the day it moved; unchanged prices keep their old date rather than claiming to be fresh. Every price change is kept in the site's committed history, which is what the “Changed” column and the change percentages read. Last update: .

What the numbers do not cover

Machine-readable data

The same figures are published for tools and agents: catalog.json holds every model with every provider offer, matrix.json the provider × model matrix, and llms.txt an index for LLM crawlers. Model, provider and plan pages are also served as Markdown by appending .md to the URL without the trailing slash.

How to cite Sovyron

Cite as: Sovyron (https://sovyron.com/), LLM API prices as of 2026-10-04. Link the specific model or provider page you used so the figure can be checked against the current table.