Engram - Qwen3-8B - FinQA: leaderboard
Engram v1 S5.T2.9, native task score (%), full reported method roster. Student backbone Qwen3-8B; parameter adaptation and trace-graph retrieval methods remain explicit configurations. Do not transfer adapted scores to the frozen base checkpoint. Source confidence intervals and exact row identities pinned. Held-out performance only, not resource efficiency or admission/retention rates.
Metric: Accuracy / task score (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3-8B (Engram Run: task=FinQA;method=APOLLO;effort=source-unreported) | 70.9 | |
| 2 | Qwen3-8B (Engram Run: task=FinQA;method=Engram;effort=source-unreported) | 69.3 | |
| 3 | Qwen3-8B (Engram Run: task=FinQA;method=MAGE;effort=source-unreported) | 67.5 | |
| 4 | Qwen3-8B (Engram Run: task=FinQA;method=FinQANet;effort=source-unreported) | 62.2 | |
| 5 | Qwen3-8B (Engram Run: task=FinQA;method=CoT;effort=source-unreported) | 57.5 |
Interactive version: theaggregate.ai/benchmark?slug=engram-qwen3-8b-finqa · How It Works · Data refreshed daily, snapshot 2026-11-12.