Independence
Rankings are not for sale
VibeDex is an independent comparison engine that runs its own blind benchmarks. We are not an aggregator, and we are not affiliated with Artificial Analysis or any lab. Because we also do paid research work, the rules separating money from rankings are written down here, in public, where they can be held against us.
By Johnathan Kwok · Founder, VibeDex
Last updated: July 2026
Can a company pay to change its VibeDex Score?
No. Scores, ranks, verdicts, inclusion, and exclusion are never for sale, never previewed for approval, and never edited for a client. There is no price, package, or partnership tier that touches benchmark output.
This is the first rule because everything else depends on it. If a prospective client asks for influence over a public score as a condition of working together, we decline the engagement.
What can vendors actually pay for?
Private synthesis and labor, never outcomes. A paid readout gives a vendor the full evidence detail behind their own public scores, a competitive gap analysis, and a working session. It includes no veto, no edits, and no influence on what we publish.
Paid engagements also feed back into public work: every private engagement yields at least one public artifact, such as an anonymized statistic or a landscape map. We do not do dark consulting.
How do free accounts and sandbox access work?
We accept reviewer accounts and sandbox access from any vendor, because benchmarking paid-tier capabilities accurately requires paid-tier access. Access-in-kind buys coverage accuracy and nothing else. It does not affect scores, framing, or timing.
Vendors who provide access get the same treatment as vendors who do not: same prompts, same judges, same disclosure rules.
How does a platform or model get benchmarked?
Inclusion is free and follows the published methodology: image and video models run the same blind prompt set as every other model, and creative platforms are assessed against the same capability framework. You cannot pay to be included, and you cannot pay to be left out.
New image models are typically benchmarked within days of release. If you want your product covered, the fastest route is to get in touch and offer reviewer access.
What happens when results are unfavorable?
They publish anyway. Reviews of partner platforms carry the same trust-risk flags and weaknesses as everyone else’s, and partner status is disclosed in the ranking itself. A commercial relationship changes the disclosure line, not the result.
Where we apply an editorial adjustment to a score, the adjustment, its direction, and its rationale are disclosed on the page rather than silently applied.
Who audits VibeDex?
We audit ourselves and publish the results. Our second-judge reliability study blind re-scored the benchmark with an independent judge model: it reproduced our one disclosed editorial adjustment to within 0.03 and corroborated the tier structure, with a published judge harshness offset of 0.18.
The self-audit program is ongoing: reliability studies, red-team passes, and published ties continue regardless of commercial work, and their results publish whether or not they flatter us.
Want your product benchmarked, or a private readout?
Start with how we test, then get in touch. Reviewer access speeds up accurate coverage. Nothing you send us, including money, moves a score.