Changes,
in the open.

List prices for every benchmarked plan

Each plan now shows its monthly and hourly list price from the provider’s own catalog, captured during the tests: DigitalOcean and Vultr on 17 October and Linode on 18 October in USD, Hetzner on 17 October in EUR net of VAT, per location and including its separately billed public IPv4 address.

Prices are before tax and are not converted between currencies. They describe the catalog at test time, not a current quote.

Agent 0.2.1

Fixes Linode JSON metadata detection. Workloads and the pilot test profile are unchanged, and published reports keep the agent version they ran with.

Akamai (Linode) batch

36 reports from Ubuntu 24.04 deployments of four Linode plans (g6-standard-2, g6-standard-4, g6-dedicated-2 and g6-dedicated-4) in London 2, Fremont and Mumbai, with three separate deployments per plan and location. Tests ran between 11:41 and 13:31 UTC with agent 0.2.0 and the pilot-1 profile.

Hetzner batch

24 reports from Ubuntu 24.04 deployments of four Hetzner Cloud plans (cpx22, cpx32, ccx13 and ccx23) in Nuremberg and Singapore, with three separate deployments per plan and location. Tests ran on 17 October between 15:53 and 18:24 UTC with agent 0.2.0 and the pilot-1 profile; the reports were published on 18 October.

Internet speed tests

Network results now come from internet speed tests run on the benchmark VM itself or on a separate deployment of the same plan and location: LibreSpeed CLI 1.0.14 downloading and uploading over HTTPS and IPv4 with three streams for 10 seconds per direction, against Clouvider’s Frankfurt and New York servers, without a rate limit. Each value links to the test it came from.

The earlier capped checks against two controlled London endpoints remain in the original reports as historical path checks.

First benchmark batch: DigitalOcean and Vultr

69 reports from Ubuntu 24.04 deployments: 36 DigitalOcean and 33 Vultr. Eight plans with shared or dedicated CPU, 2 vCPU / 4 GiB and 4 vCPU / 8 GiB, in London, US West and Bangalore, with three separate deployments per plan and region. Vultr’s vc2-4c-8gb plan was unavailable in Bangalore, so no substitute was tested.

Tests ran between 11:39 and 13:02 UTC with agent 0.2.0 and the pilot-1 profile. Each report has 40 passed workloads; the 24-hour consistency check was skipped. Reports were published at 13:27 UTC. Prices were not recorded.

Pilot-1 test profile

The pilot profile covers CPU, memory read and write, disk, network and PostgreSQL in one short test window. It does not include memory copy, PostgreSQL latency percentiles or a 24-hour consistency run.

Network results use two controlled London endpoints with a 100 Mbit/s rate setting, so they do not measure maximum link capacity. Methodology v1.0 remains the broader proposed programme; each pilot report states which parts it covers.

AI-assisted evidence review disclosed

Reports in this batch received AI-assisted evidence review under user-delegated authorization. Each report names its reviewer and lists its limitations. This is not independent human verification or hardware attestation.

Sample data removed

An earlier interface preview used synthetic measurements. Those examples are no longer served: sample report URLs return “not found”, and the benchmark pages read reviewed publications only.

Agent and evidence review workflow

Open-source agent execution, private submission, complete evidence review and explicit approval of the exact report revision are required before publication.