Third-Party Review: Kimi K3 Agent Mission Nearing Fable 5, but Illusions Abound

According to Perceive Beating monitoring, Artificial Analysis has released the Kimi K3 independent assessment. It scored 57 points in the Intelligence Index v4.1. Its overall level is comparable to Claude Opus 4.8 and GPT-5.5, but still lower than Claude Fable 5 and GPT-5.6 Sol. This composite index consists of 9 tests covering real-world tasks, tool invocation, programming, reasoning, knowledge, and long contexts.

Kimi K3's Agent performance is more outstanding. It scored 1668 Elo in the real-world task test GDPval-AA v2, surpassing Opus 4.8, GPT-5.5, and GLM-5.2. Its score in the automated SaaS workflow test is 53%, ranking first. In long-process knowledge work, it ranks second, just below Fable 5.

In this index, Kimi K3 has an average cost of $0.94 per task. GPT-5.6 Sol costs $1.04, while Opus 4.8 costs $1.80. However, its illusion rate in the knowledge reliability test has increased from 39% to 51%, and the accuracy has improved from 33% to 46%.