Score the answer, not the name.

Independent evaluation of AI models. The same answer always earns the same score: no favourites, no exceptions. Every public result was placed on the record by its owner's approval.

mizan ❯ Ask me anything about this house -- scoring, waivers, submissions, the public record.

how a run goes

[submit] your weights, or your own endpoint
[verify] every uploaded file checked against its hash
[sandbox] one GPU instance for this run, no route to the internet
[load_check] the model answers the test prompts
[confirm] you see the result and say go
[generate] every question, the same settings for every model
[grade] MIZAN reads each answer blind, by its meaning
[score] the same answer always earns the same score
[report] what it does well, where it fails, what it needs
[publish] only if you approve it
[submit] your weights, or your own endpoint
[verify] every uploaded file checked against its hash
[sandbox] one GPU instance for this run, no route to the internet
[load_check] the model answers the test prompts
[confirm] you see the result and say go
[generate] every question, the same settings for every model
[grade] MIZAN reads each answer blind, by its meaning
[score] the same answer always earns the same score
[report] what it does well, where it fails, what it needs
[publish] only if you approve it

playground$

arena$

operator-4b$

step 1$

step 2$

step 3$

REGISTERED ACCESS ONLY

RECORDED WAIVER FIRST

SECURITY AND PRIVACY BY DEFAULT

OWNER-APPROVED PUBLICATION