The new Kimi-K3 model recorded a preliminary ECI score of 155.53, with a 90% confidence interval of 153.87–158.21. This performance places Kimi-K3 slightly ahead of Anthropic's Opus 4.6, which scored 155.31, and OpenAI's GPT-5.3-Codex, which achieved 155.36.
A new model from an emerging lab is demonstrating competitive performance against frontier models, signaling a broadening competitive landscape.