China's Kimi K3 AI model matches OpenAI's mid-tier GPT-5.6 Terra at lower cost

Moonshot AI's Kimi K3 model has demonstrated capabilities 'extremely close' to OpenAI's flagship, according to Aikido researcher Philippe Dourassov. In a private test against 26 known vulnerabilities, Kimi K3 discovered 23, matching OpenAI's GPT-5.6 Terra while running at a quarter of the cost of the Sol model. Dourassov noted that because the test used recently discovered vulnerabilities, Kimi K3 could not have trained on them, indicating a genuine leap in performance. This development fuels concerns that US safety regulations are hindering domestic AI progress, as Chinese open-source models rapidly close the gap.