😱 ScreamingFace
draft Fusion scores are internal results (Aug 12) pending publication; solo baselines are illustrative until the methodology page ships. For review only.
the verified fusion leaderboard

Weld two models. Beat the best one.

Fusions — combinations of the models you already pay for — score higher than any single one of them. Every score on the board ships with its recipe and re-runs in one line. We train nothing; the gain comes from joining models that already exist.

See the board → Fusion Monsters — apply →
DRACO · open-context news benchmark · internal results, Aug 12
five-model fusion · ours68.6%
open-source fusion · ours61.9%
best single model *58.2%
budget flagship *52.4%

* solo bars illustrative pending the methodology page · verified = we re-ran the recipe · the full board →

≥ $1.78/task Above that price, the fusion frontier sits above the solo frontier — more accuracy for the same spend. Below it, a single small model is still the right call. The board shows both, so you can pick by budget instead of by brand.

The loop: pull the recipe closest to your task · change one variable · run it against the benchmark · publish. Your win goes up under your name — and becomes the next person's starting point.

the community program

Fusion Monsters

The hunt for the next verified score. Selected applicants get $50 of challenge credits on open models, a baseline we set with our own strong attempt, and a deadline. Beat it and you're in: a monthly allowance, the supported benchmarks, and a board where your win carries your name — until someone forks your recipe and beats you. 😱

allowance $300/mo admissions rolling · no rejections winning recipes published, under your name
Apply — takes 5 minutes →

Get the next one

One email per verified result. Nothing else.

😱 ScreamingFace — the verified fusion leaderboard