benchr
Sources and evidence types
Official pricing, specifications, release dates, and deprecation claims are sourced from provider documentation. Third-party benchmark results are sourced from the benchmark publisher or maintainer. benchr estimates and analytical judgments are labeled as editorial estimates or analysis. A claim without sufficiently reliable support is removed or marked as uncertain.
Independence
benchr does not sell editorial rankings, favorable conclusions, or paid placement disguised as reporting. Commercial relationships do not determine which models are covered or how they are evaluated, and partners do not preview, approve, or edit coverage. That is what editorial independence means here; it does not mean the publication receives no compensation.
Revenue and commercial disclosures
Some links to AI/ML API, Lovable, and RunPod are affiliate links. benchr may receive a commission when a reader uses a marked link, at no extra cost to the reader. Affiliate links are disclosed near the recommendation and identified to search engines with rel="sponsored" where applicable. Google AdSense is planned only after approval and is not currently loading ads. See the affiliate disclosure for the current programs and the privacy policy for the current advertising status.
Verifiability
Numeric items should identify their evidence type. Official pricing links to the provider's pricing page. Benchmark scores link to the benchmark maintainer — including LMSYS Arena, SWE-bench Verified, ARC-AGI, and SimpleBench. Release dates link to the provider's announcement. A benchr estimate or analytical judgment is labeled as such and is not presented as an official or third-party benchmark result.
Authorship and review accountability
Articles are attributed to the benchr Editorial Team, the publication-level Organization responsible for scope, evidence checks, language editing, release checks, and corrections. These are responsibilities, not a claim that benchr employs a separate person for each role. A personal byline requires a real public name, role, profile URL, and explicit article assignment; no identity or credential is inferred.
A “Reviewed” dateline records a source-check event under this process. It does not establish a second reviewer. Structured data does not claim reviewedBy without a separate, truthful public reviewer record.
Use of AI in production
AI tools can assist with outlining, draft language, translation checks, code, and custom editorial illustrations. Their output is never accepted as evidence, and the tool is not credited as an author or independent reviewer. A price, specification, benchmark result, model ID, or date needs a linked provider or benchmark source before publication; an unsupported value is removed or shown as unavailable. The publication, not the tool, remains responsible for source selection, wording, approval, and corrections.
Generated illustrations use benchr's own visual system and are not official provider artwork. They must not reproduce a provider logo, imply endorsement, or place a factual claim inside an image that is absent from the article. An image is decoration or explanation, never proof of a model capability.
Conflicts of interest and access
A commercial relationship does not change a rating, ordering, or conclusion. If a provider supplies paid access, credits, a review unit, or material non-public access for a specific article, that fact must be disclosed beside the coverage. Partners do not receive approval rights. When the evidence is too thin to support a verdict, the page should say so rather than convert access or an affiliate relationship into a recommendation.
What is and isn't covered
In scope: pricing, benchmarks, capability comparisons, context windows, deprecation status, use-case recommendations, and analysis of the closed and open model markets.
Out of scope: AGI and AI safety policy debates, model-training research, AI as a social phenomenon. Standalone reviews of dedicated image- or video-generation services (Midjourney, Sora, etc.) are also out of scope. Use-case guides covering image and video generation as a feature of text-first models (GPT, Claude, Gemini) are in scope.
Corrections
See the corrections page for a list of material corrections and the article-level changelog at the bottom of each piece for granular history.