How Much Do LLMs Hallucinate in Document Q&A Scenarios? A 172-Billion-Token Study Across Temperatures, Context Lengths, and Hardware Platforms

TL;DR

RIKER, a ground-truth-first evaluation methodology that enables deterministic scoring without human annotation, reveals that models that excel at finding facts may still fabricate facts that do not exist and confirms that deployment decisions need not be hardware-dependent.

Abstract

How much do large language models actually hallucinate when answering questions grounded in provided documents? Despite the critical importance of this question for enterprise AI deployments, reliable measurement has been hampered by benchmarks that rely on static datasets vulnerable to contamination, LLM-based judges with documented biases, or evaluation scales too small for statistical confidence. We address this gap using RIKER, a ground-truth-first evaluation methodology that enables deterministic scoring without human annotation. Across 35 open-weight models, three context lengths (32K, 128K, and 200K tokens), four temperature settings, and three hardware platforms (NVIDIA H200, AMD MI300X, and Intel Gaudi 3), we conducted over 172 billion tokens of evaluation - an order of magnitude beyond prior work. Our findings reveal that: (1) even the best-performing models fabricate answers at a non-trivial rate - 1.19% at best at 32K, with top-tier models at 5 - 7% - and fabrication rises steeply with context length, nearly tripling at 128K and exceeding 10% for all models at 200K; (2) model selection dominates all other factors, with overall accuracy spanning a 72-percentage-point range and model family predicting fabrication resistance better than model size; (3) temperature effects are nuanced - T=0.0 yields the best overall accuracy in roughly 60% of cases, but higher temperatures reduce fabrication for the majority of models and dramatically reduce coherence loss (infinite generation loops), which can reach 48x higher rates at T=0.0 versus T=1.0; (4) grounding ability and fabrication resistance are distinct capabilities - models that excel at finding facts may still fabricate facts that do not exist; and (5) results are consistent across hardware platforms, confirming that deployment decisions need not be hardware-dependent.

Figures (2)

• Figure 1: RIKER methodology overview. Traditional approaches (left) extract ground truth from existing documents via expensive human annotation, producing static benchmarks vulnerable to contamination and requiring biased LLM judges. RIKER (right) inverts this: structured ground truth is defined first, then documents and questions are generated from it, enabling deterministic scoring and regenerable corpora. (Figure reproduced from roig2026scalablereliableevaluationai.)
• Figure 2: Grounding accuracy vs. fabrication rate at 32K context. Each model is shown at its independently optimal temperature for each metric. The dashed line shows the expected inverse relationship; deviation above the line indicates excess fabrication. Color and shape indicate tier: well-calibrated (green circles, Gap $<$ +5), mid-tier (orange squares), and fabrication-prone (red triangles, Gap $\geq$ +15).