Work with us.
Codesota is a one-person registry project, maintained in Warsaw. It is currently self-funded. It has no paid advisory board, no sponsored rankings, and no placement-for-hire slots. Those things are not on the roadmap.
What I do take on — sparingly and with the terms written down — is compute support for reproductions, commissioned evaluations, and consulting engagements around model selection. If any of that is what you are looking for, this page tells you how it works.
What the work is.
Three categories. Anything outside of these is almost certainly a no — polite, but a no.
GPU hours or API credits to run reproductions on the public registry. In return your infrastructure is credited on the reproduction notes for the runs it powered — nothing more, nothing ranked.
Evaluations outside the normal editorial queue — a private eval, a domain your team cares about, a custom harness. Deliverable is the methodology and the numbers; you pay for time, not for placement.
Short engagements helping a team choose between models, design an internal benchmark, or read an RFP critically. Hourly, scoped, and written down.
Rules of the house.
Editorial order of the public registry is determined by reproducible evidence only. No partner, past or future, can change the top row of a table by paying for it. This includes highlight slots, sort defaults, and badge colours.
When a page, a benchmark, or an evaluation was paid for, that fact appears on the page in the same voice as the rest of the text — not in small grey type at the bottom.
Anything I measure as part of a commissioned benchmark is published under the same open-data terms as the rest of the registry. You do not get to buy the only copy.
Any NDA is limited to unpublished internal numbers you share with me during the work. It does not cover the public results, the methodology, or the existence of the engagement itself.