Artificial Analysis scores Muse Spark 1.2 at 54 and traces its lower hallucination rate to abstention rather than accuracy

Sources: benchmarks, discussion
Summary: Artificial Analysis scores Muse Spark 1.2 at 54 on its Intelligence Index. On AA-Omniscience the score rises from 18 to 22 while the hallucination rate falls from 38% to 28%, the attempt rate drops from 82% to 67% and accuracy falls from 41% to 38%. Cost per Intelligence Index task rose from $0.29 to $0.40 at unchanged pricing of $1.25 and $4.25 per 1M input and output tokens, which the article attributes to token usage rather than price.
Why it matters: The lower hallucination rate comes from answering less rather than knowing more, so the model behind today's Muse Code release declines more questions instead of getting more of them right.