Which AI hallucinates the most?
Asked "Which AI hallucinates the most?", ChatGPT named 17 distinct names across 10 answers on September 8, 2026, and 9 of them in two or more answers.
Measured acrossChatGPT
9 of 17 names confirmed · named in 2 or more of 10 answers · asked September 8, 2026 · 1 engines
The AI model that hallucinates the most depends on the task, benchmark, and model version, but smaller, less fine-tuned models tend to hallucinate more than larger, retrieval-augmented systems.
- 1OpenAInamed in 4 of 10 answers
- 2Googlenamed in 3 of 10 answers
- 3Anthropicnamed in 3 of 10 answers
- 4GPT-3named in 2 of 10 answers
- 5Claudenamed in 2 of 10 answers
- 6LLaMAnamed in 2 of 10 answers
- 7GPT-4named in 2 of 10 answers
- 8Gemininamed in 2 of 10 answers
- 9Metanamed in 2 of 10 answers
- 10ChatGPTnamed in 1 of 10 answersone answer
- 11DeepSeeknamed in 1 of 10 answersone answer
- 12Perplexitynamed in 1 of 10 answersone answer
- 13Claude 3.5named in 1 of 10 answersone answer
- 14Groknamed in 1 of 10 answersone answer
- 15PaLMnamed in 1 of 10 answersone answer
- 16BioGPTnamed in 1 of 10 answersone answer
- 17MedPaLMnamed in 1 of 10 answersone answer
LOCKED
The full measurement
- The position each of the 1 engines gave all 17 names.
- How many of the 10 answers named each of them.
- 4 sampled observations behind this ranking, and where the engines disagree.
- Fan-out — the query each engine actually searched.
- Every citation, and the sources nobody cited.