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

TL;DR

This study quantifies hallucination rates in large language models during document question answering, revealing that fabrication increases with context length and varies by model, temperature, and hardware, using a novel deterministic evaluation method.

Contribution

Introduces RIKER, a ground-truth-based evaluation approach for deterministic, scalable measurement of hallucinations in LLMs across diverse models and hardware platforms.

Findings

Hallucination rates increase with context length, reaching over 10% at 200K tokens.

Model choice significantly impacts hallucination resistance, more than size or temperature.

Grounding ability and fabrication resistance are distinct, with some models excelling at one but not the other.

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.