the verified fusion leaderboard · GPQA Diamond
Fusion beats S
TA
Every other leaderboard ranks one model at a time. We rank fusions — and a verified fusion of frontier models beats the best single model on the board. Every score is run in a sealed enclave and ships with a receipt you can check.
best single model
84.2%
GPT-5.5, alone
+4.2
verified fusion · SOTA
88.4%
three models, voting, enclave-verified
leaderboard · GPQA Diamond
😱 screamingface fusion88.4%
GPT-5.584.2%
Claude Opus 4.883.1%
Gemini 3 Pro82.0%
| # | configuration | accuracy | questions | verified |
|---|---|---|---|---|
| 1 | 😱 screamingface fusion | 88.4% | 198 | ✓ enclave receipt |
| 2 | GPT-5.5 | 84.2% | 198 | ✓ enclave receipt |
| 3 | Claude Opus 4.8 | 83.1% | 198 | ✓ enclave receipt |
| 4 | Gemini 3 Pro | 82.0% | 198 | ✓ enclave receipt |
why you can trust this
Every run executes in a sealed enclave and produces a signed certificate — the same eval, on
the same questions, reproducible by anyone. It's the only registry where the number comes with
a receipt, not a press release.
get the next one
We knock down a new benchmark on a regular cadence. Get each verified result the day it lands.
No spam. One email per new verified leaderboard.
😱 screamingface
verified fusion leaderboard