OFFER VALUE · 2026-09-21.value-v2

Value calculated from the offering.

Every offering has a score from 0 to 100. It uses the values available in the catalog and recalculates when those values or your usage assumptions change. Missing fields lower confidence or broaden the comparison; they do not prevent a score.

The price formula

Score = round(100 × reference price ÷ (reference price + offering price))

The reference is the median positive price in the comparison group, with each provider contributing its cheapest eligible offering once. An offering at the reference price scores 50; one at half that price scores 67. Zero modeled cost scores 100. For example, a reference of $120 and an offering price of $90 produce 57/100. This is a relative value score, not a review grade or a performance benchmark.

Use the available data

First, we calculate the bill from the listed rate, usage, minimum purchase, billing period, setup and required fees. If some billing details are missing, we use an estimated cost from the available listed price and label its confidence low. The complete bill remains separate from that estimate. Missing prices never become zero.

We first look for another provider in the same category with the same chip and RAM, then widen to the same chip, the same RAM, or the category as a whole. CPU, storage and every other field do not need to be filled in. There is no minimum of three providers. Prices in different stated currencies are kept separate, with no exchange-rate conversion. Unconfirmed currency notation is grouped separately and has low confidence.

If a usable price is absent, the provisional score averages the available CPU, RAM and storage capacity scores. Each is 100 × documented quantity ÷ (documented quantity + category reference), using the median of each provider’s median quantity as the reference. CPU kinds and RAM units are kept separate. This estimates capacity, not performance per dollar. When no usable price or capacity quantities are available, the offering gets a provisional neutral 50/100. Its source gaps remain visible.

Usage and commitments

CI defaults to 1,000 compute minutes per month over 12 months, in 10-minute jobs, with private repositories and one user. Dedicated Macs and VMs default to one continuous 720-hour allocation. Change the assumptions above the table. Minutes represent compute consumption, not completed builds.

Known billing periods, job increments and minimum allocations feed the cost. A two-hour rental with a 24-hour minimum pays for 24 hours. Annual purchases include the full payment even during a shorter comparison. Alternative billing terms are separate purchases; the lowest modeled bill in the selected billing unit and currency is used. Calendar estimates use 30-day months, 90-day quarters and 365-day years. Optional add-ons, transfer, extra storage and applicable taxes are outside the base scenario.

Required per-user plans use one user. Shared account credits are assumed already used. Known bundle overage prices are included. If a paid bundle has no documented overage price, a provisional estimate may extrapolate its unit cost; this is labeled and is not an available purchase quote. Exhausting a free bundle never makes additional usage free. Unresolved fees, credit purchase rules and older evidence remain visible beside the estimate.

Included usage and Apple membership

Eligible zero-cost usage receives 100/100 and the label “Free / included.” Public-repository eligibility must be documented for the specific offering. Paid larger runners do not inherit another runner’s free allowance.

When Apple membership is already held, the calculator excludes that existing cost. When a new membership is needed, it includes the full annual fee from the catalog, payable upfront. A paid compute bundle replaces its plan allowance; membership hours are not added again. No provider receives bonus points based on its name.

Confidence is separate

High confidence means a modeled cost with current billing evidence and either documented zero cost or another provider matching chip and RAM. Medium confidence means a modeled cost with a broader comparison. Low confidence covers partial cost estimates, unconfirmed currency, a single-provider reference and capacity or neutral fallbacks. Confidence describes the inputs and comparison, not measured hardware performance.

Region, software, services and capability filters narrow the comparison group. Text search, provider selection, sorting and pagination do not change the reference group. Equal specifications do not establish equal speed, isolation, uptime or support. Availability is separate from value; missing inventory is not zero stock.

Sorting and updates

“Offer value: high to low” sorts every offering by its numeric score. The default discovery order still spreads providers across published-price groups. Open “Calculation & assumptions” for the score basis, reference prices, known charges and sources. Published catalog changes automatically feed the formula; the formula itself does not fetch external pricing pages.

Historical provider service assessments remain in compatibility API fields and do not contribute to the current offering score.