Gemini 3 Flash scores 71, only two points under Gemini 3 Pro at half the price, but uses over 2x the tokens of its predecessor and hallucinates 91% on hard questions.
Key takeaways · AI-distilled
In December 2025, Artificial Analysis found Gemini 3 Flash Preview improved knowledge accuracy but had a 91% AA-Omniscience hallucinationWhen a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.Full definition → rate. Leading that combined benchmarkA standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.Full definition → did not establish reliable abstention.
Gemini 3 Flash Preview used about 160 million output tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → in the December 2025 Intelligence Index tests, over twice Gemini 2.5 Flash. Artificial Analysis attributed its cost efficiency to low token prices despite that increased usage.
Artificial Analysis reported text, image, video and audio inputs for Gemini 3 Flash Preview in December 2025. Its cited MMMU-Pro result tested image reasoning specifically, so it did not establish performance across every supported input type.
Terms in this piece · Glossary
multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.
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.
hallucination — When a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
