Engram - Qwen3-8B - HotpotQA F1: leaderboard
Engram v1 S5.T2.7.2, native task performance (%), full reported method roster. Student backbone Qwen3-8B; trained adapters and memory frameworks remain native run configurations, not released-model aliases. Full table and confidence intervals retained; no resource, conditional verification or retention rate admitted.
Metric: Accuracy / task score (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3-8B (Engram Run: method=Engram;effort=source-unreported) | 71.5 | |
| 2 | Qwen3-8B (Engram Run: method=EmbQA;effort=source-unreported) | 55.8 | |
| 3 | Qwen3-8B (Engram Run: method=EvolveR;effort=source-unreported) | 54.6 | |
| 4 | Qwen3-8B (Engram Run: method=SuRe;effort=source-unreported) | 53.5 | |
| 5 | Qwen3-8B (Engram Run: method=RPG;effort=source-unreported) | 49.2 | |
| 6 | Qwen3-8B (Engram Run: method=RoleRAG;effort=source-unreported) | 49.2 | |
| 7 | Qwen3-8B (Engram Run: method=ReAct;effort=source-unreported) | 41.3 | |
| 8 | Qwen3-8B (Engram Run: method=CoT;effort=source-unreported) | 39.8 |
Interactive version: theaggregate.ai/benchmark?slug=engram-qwen3-8b-hotpotqa-f1 · How It Works · Data refreshed daily, snapshot 2026-11-12.