Independent Assessment
Nomi AI Reliability Report
An independent, ongoing assessment of major AI sources rated across four reliability pillars: Factual Accuracy, Internal Consistency, Source Traceability, and Bias & Balance. Updated monthly.
Scores are derived from structured Nomi evaluations of AI-generated outputs across diverse factual query categories. Each source is evaluated on a rolling basis. Scores reflect aggregate performance — individual results may vary by query type and deployment context.Loading live data…
Sources Evaluated
6
Average Score
69
High Assurance
1 / 6
Top Performer
Perplexity
Overall Score Comparison
Perplexity AI
Perplexity leads on source traceability due to its inline citation model. Live retrieval reduces hallucination risk significantly, making it the most verifiable general-purpose AI source evaluated.
Claude 3.5 Sonnet
Claude 3.5 Sonnet excels at consistency and balance, making it reliable for analysis tasks. Its lack of citation capability is the primary limitation for assurance-critical use cases.
ChatGPT (GPT-4o)
GPT-4o performs well on general factual queries but falls short on source traceability. Suitable for drafting and reasoning tasks where outputs will be independently verified.
Gemini 1.5 Pro
Gemini 1.5 Pro benefits from Google Search integration, improving factual grounding on current events. Source traceability remains partial — citations are present but not always verifiable.
Microsoft Copilot
Copilot provides web-grounded responses but citation quality is inconsistent — links often point to aggregator pages rather than primary sources. Multi-turn consistency is a known weakness.
Meta Llama 3.1
Llama 3.1 scores lower primarily due to deployment variability — the open-source model's reliability depends entirely on how it has been fine-tuned and deployed. Auditability of weights is a genuine strength.
Evaluation Methodology
How scores are calculated
Each AI source is evaluated using Nomi's four-pillar framework across a standardized set of factual queries spanning science, law, medicine, history, and current events. Scores represent the aggregate of all evaluations in the current cycle.
Rating thresholds
HIGH (75–100): Consistently meets reliability standards.
MODERATE (50–74): Generally sound with notable gaps.
LOW (25–49): Significant reliability concerns.
INSUFFICIENT (<25): Does not meet minimum standards.
Want to run your own evaluation? Try Ask Nomi →
Baseline data · Updated July 2026