How we score
Every score uses the same rules and only sourced facts. Version 1.3 (2026-11-14). Vendors cannot pay to change a score, and user reviews are shown next to the score, not mixed into it.
The parts
| Part | Max points | Rule |
|---|---|---|
| Price | 30 | Comparable list price against the category median, on a log scale: the median price gets 15, half the median 22.5, a quarter or less 30, double the median 7.5, four times or more 0. Code sandboxes, dev boxes and CI runners use a monthly cost for a reference workload, with paid plan fees and the usage those plans include. Free allowances are not subtracted: they score in Free tier. A fixed-size product whose vCPU count is not published gets no comparable price. Browser sandboxes use the price per browser-hour, GPU sandboxes the price of one H100 per hour. Coding agents are not price-ranked. Open-source or self-hostable software with no hosted list price gets the median price points of its category (15 of 30), and its card says "Self-host, no list price". |
| Free tier | 10 | 10 for free use every month (including free software), 6 for a one-time credit or trial, 0 for none. |
| Isolation | 15 | microVM 15, virtual machine 15, sandboxed container (gVisor or similar) 11, V8 or WebAssembly isolate 9, container 6, OS-level process sandbox 4, not stated 3. "Not stated" gets the lowest points: we do not guess. |
| Persistence | 10 | 5 for pause and resume with state, 5 for snapshots. |
| Limits | 10 | Session length: 4 for 24 hours or more, 2 for 1 hour or more. Max vCPU: 3 for 8 or more, 2 for 4 or more. Max memory: 3 for 16 GiB or more, 2 for 8 GiB or more. Limits the vendor does not state get 0. |
| Docs and SDKs | 10 | 3 for public docs, plus 5 for three or more SDKs or interfaces (4 for two, 2 for one), plus 2 for open source or self-hosting. |
| External benchmarks | 15 | Code sandboxes and dev boxes only: median start time in the weekly ComputeSDK burst test (100 sandboxes at once), counted only when 95% or more of the starts succeed. Absolute log scale: 100 ms or faster gets 15, 5 seconds or slower gets 0 (1 second gets about 6). A provider that is not in the benchmark gets the median points of the measured providers in its category, so not being measured neither helps nor hurts. Its page says "speed not measured". We do not run benchmarks ourselves. |
Missing data
When an input does not apply or is not available (for example, no public price), that part is left out and the score is the share of the remaining points. A missing benchmark gets the category median, never zero. A fact the vendor does not publish counts as zero, because buyers cannot rely on it.
Changes to the method
- 1.3 (2026-11-14). "Self-hosted" is no longer a category. Its 17 entries are rated in their real category (mostly code sandboxes; some dev boxes and browsers), and open source and self-hostable are filters. Software with no hosted list price gets the median price points of its category instead of having the price part left out.
- 1.2 (2026-11-13). Code sandboxes and dev boxes that are not in the start-time benchmark now get the median benchmark points of the measured providers. Before, that part was left out for them, so a slow measured provider could lose points that an unmeasured one never risked.
- 1.1 (2026-11-13). Price now uses list price: free monthly allowances no longer lower the comparable price, because the Free tier part already scores them (one provider showed $0 a month). Price points now come from the distance to the category median on a log scale, not from rank, so two prices a few cents apart get almost the same points. Benchmark points now use an absolute start-time scale, not rank among the few measured providers, so being measured is not a penalty in itself. A fixed-size product with no published vCPU count no longer gets a monthly estimate.
- 1.0 (2026-11-13). First published version.
Sources and checks
- Every number comes from the vendor's own pricing page, docs or repository, with the URL, an exact quote and the date we checked it. Each provider page lists them.
- We read every pricing page daily and check that each quoted price is still on it. A few pages that only load in a full browser are checked weekly. A changed page is re-read and the change is logged on price changes. When a page cannot be read, the provider shows "price check failed" with the date, and its prices keep the date they were last confirmed.
- We list providers that are live, open to self-serve sign-up, and publish pricing or docs. Open-source projects need a license and a release or commit in the last six months.
User reviews and upvotes
Only signed-in users can upvote or review, once per provider. We never write, buy or seed reviews or votes, and we do not offer anything in return for them. We remove only spam, abuse, or reviews by people connected to the vendor or a competitor.
Money
Listings are free. If we add paid profiles or sponsored placements later, they will be labelled "Sponsored", their links will use rel="sponsored", and they will never change a score or hide a review.