INTELLECT-3: benchmark results
Prime Intellect's open 106B A12B reasoning MoE, post-trained from GLM-4.5-Air-Base using supervised fine-tuning and reinforcement learning. Provider: PrimeIntellect. Released 2025-11-26. Access: Open.
Aggregate Rating 1604 ± 18, rank #427 of 1656 rated models, from 93 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| IKP - T7 | 15.7 | Accuracy (%) | 86.8 |
| AA Omniscience - Software Engineering (SWE) - Dart | 26 | Accuracy (%) | 77.6 |
| Medmarks - MedQA | 90.52 | Score (%) | 77.3 |
| Medmarks - MedHallu Easy | 70.67 | Score (%) | 75 |
| Medmarks - MedConceptsQA Medium | 74.1 | Score (%) | 73.9 |
| Medmarks - MedConceptsQA Easy | 97.95 | Score (%) | 72.7 |
| UGI Leaderboard | 42.24 | UGI Score | 71.9 |
| Medmarks - MetaMedQA | 75.31 | Score (%) | 71.6 |
| UGI - Natural Intelligence | 29.75 | NatInt Score | 71.6 |
| Medmarks - HEAD-QA v2 | 88.3 | Score (%) | 70.5 |
| AA Omniscience - Software Engineering (SWE) - Swift | 40 | Accuracy (%) | 70.1 |
| Medmarks - M-ARC | 52.33 | Score (%) | 69.3 |
Interactive version: theaggregate.ai/model?slug=intellect-3 · How It Works · Data refreshed daily, snapshot 2026-10-20.