ProKope-AI/ProKope-421M
Text Classification • 0.4B • Updated • 23 • 1
None defined yet.
Replacing slow, expensive autoregressive cloud generation with single-forward-pass deterministic decision models. Sub-25ms latency, zero output tokens, and 71.40% benchmark accuracy.
Experience zero-token non-autoregressive decision classification in real time.
Witness why 20ms System 1 decision models can pilot high-speed autonomous agents where cloud LLMs crash.
Click any lane on the road to manually spawn hazard roadblocks.
Empirical performance across 400 test cases and 2,000 calibrated decisions.
| Model Architecture | Parameters | Mode | Accuracy ↑ | Brier Score ↓ | p50 Latency |
|---|---|---|---|---|---|
| Featherless Simple Jev (Cloud) | 35,000M (35B) | General (Zero-Shot) | 71.60% | 0.110 | ~1,200 ms |
| 🥇 ProKope-421M | 421M (0.42B) | Specialist (Sovereign) | 71.40% | 0.071 | <25 ms |
| prima-ratio + 12B (Cloud) | 12,000M (12B) | General (Zero-Shot) | 70.20% | 0.125 | ~700 ms |
| oscar-1-400m | 400M (0.40B) | Specialist | 70.00% | 0.062 | ~35 ms |
| Bekko System One v0 (Benchmark Baseline) | 400M (0.40B) | Specialist | 66.80% | 0.113 | 808 ms |
| ModernBERT-base (Benchmark Baseline) | 149M (0.15B) | Specialist | 64.60% | 0.119 | 349 ms |