Source priority
We prefer provider websites, official documentation, official repositories, and model cards. Third-party material is contextual and labeled.
Public method
Editorial and benchmark methodology
These rules cover models, workflows, integrations, templates, comparisons, and benchmarks so readers can distinguish facts, claims, and unknowns.
Independent and provider-reported results
Independent results require a public method and reproducible conditions. Provider-reported results stay labeled and are not blended into an independent ranking.
Unknown information
Fields without reliable source support remain unknown. We do not infer pricing, licenses, context limits, or performance from similar products.
Updates and corrections
Every record carries a verification date and primary sources. Readers can submit corrections for editorial review.
Benchmark eligibility rules
- The prompt, task definition, and evaluation set must be public or reproducible.
- The exact model or API version must be recorded.
- Hardware, region, concurrency, and sampling settings must be disclosed for latency claims.
- Accuracy metrics must include sample size and scoring method.
- Provider-reported results must be labeled and kept separate from independent results.
- High-impact domains require domain-specific safety review before publication.