Kimi K3 Achieves 3 In The Artificial Analysis Intelligence Index Comparable To Opus 4 8 And Gpt 5 5
Source
artificialanalysis.ai
Author
Artificial Analysis articles
Date
Why it matters
Kimi K3 reaches Opus 4.8-class results at about half the cost per task, and would lead open weightsA model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.Full definition → models if released. Its output price rose to $15 per 1M tokens, so check cost before swapping it in.
Key takeaways · AI-distilled
In its July 2026 tests, Artificial Analysis found Kimi K3 used 21% fewer 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 → than K2.6 while improving its Intelligence Index score. Higher capability came with lower output-token use in this comparison.
Artificial Analysis reported in July 2026 that Kimi K3 improved AA-Omniscience accuracy from 33% to 46%, but its hallucinationWhen a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.Full definition → rate rose from 39% to 51%. Better factual accuracy did not mean better abstention.
At Kimi K3 launch in July 2026, Moonshot had announced plans to release weights but had not released them. Artificial Analysis therefore described its prospective open-weights leadership conditionally.
Terms in this piece · Glossary
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.
open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
hallucination — When a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.
