Evidence comes first

A number becomes useful when you know where it came from.

Sources and prices

Endpoints, model IDs, prices and limits are checked against public documentation and catalogues. Each article lists sources and a consultation date. A price is a snapshot, not a promise about future costs.

Independent editorial tests

Tests published as independent must be run and documented by Deepweight, separately from figures supplied by a provider. A reproducible protocol includes model ID and version, endpoint, date, dataset, sample size, parameters, concurrency, evaluation criteria and logs with credentials removed.

No authenticated API benchmarks conducted by the editorial team are published at launch. The initial guides contain documentary analysis, implementation examples and arithmetic scenarios. They are not presented as measured hands-on results.

Communicated figures

For Uncensured AI, the publisher received figures of 250–300 tokens/s on average, depending on the model, and a 0% refusal rate. No verifiable sample and raw logs accompany them. They are identified as communicated figures, not independent Deepweight results. An observed zero does not guarantee the same result for every request.

Calculations and code

Cost scenarios identify volumes, rates and formulas. Code is an implementation example: where credentials or an execution environment are missing, no end-to-end test is claimed. Demonstration output is never attributed to a model that was not actually called.

Editorial selection

Topics and providers are selected by the publisher. A service appearing in a guide does not establish superiority across the market. A comparison needs alternatives, a task and evaluation criteria; otherwise it remains a guide to making a choice.

Corrections

Material corrections update the content and modification date without backdating publication. Send corrections to r1.rete@tutamail.com.