A small model can avoid many wrong answers by declining questions it cannot answer. Artificial Analysis' MiniCPM5-2B evaluation found exactly that pattern: relatively frequent abstention accompanied low factual accuracy. A favorable non-hallucinationWhen a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.Full definition → measure therefore did not mean broad knowledge.
The report places the model's stronger results in the context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition → of its small total parameter count, which matters for memory-constrained deployment. Those size-relative gains coexist with limits on difficult knowledge, coding and document tasks. Reading both coverage and error behavior prevents a cautious small model from being mistaken for one that can reliably answer a much wider range of questions.
Terms in this piece · Glossary
hallucination — When a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
reasoning model — A model trained to think — generating extended internal reasoning before answering — trading time and tokens for accuracy on hard problems.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
