Research
Hypereum publishes the parts of its research that let other people check its claims: model weights, benchmark data, the raw predictions of every model it compared, and the scoring code. It does not publish the training pipeline: data generation, labelling and hyperparameters.
| State | Title | Kind | Date |
|---|---|---|---|
| Provisional | HivemindEval v2.1 | Model | 2026-07-17 |
| Current | HivemindEval Benchmark, public subset | Dataset | 2026-07-17 |
| Current | HivemindEval technical report v1.0 | Paper | 2026-07-17 |
| Current | hivemind-eval, scoring code | Software | 2026-07-17 |
| Withdrawn | HivemindEval v2.0 It gave every input the same score, 75. | Model | 2026-05 |