The new Chinese model Kimi K3, from startup Moonshot, caused an earthquake in global markets last Friday. With 2.8 trillion parameters, the K3 rivals the most advanced models from OpenAI and Anthropic, surpassing GPT-5.6 and Opus 4.8 in benchmarks such as Program Bench and SWE Marathon. Bloomberg reported that the launch was dubbed a ‘new DeepSeek moment’, sending AI and semiconductor stocks down.
Analyst Ranjan Roy from Margins calls the event ‘massive’ and compares it to DeepSeek’s impact. ‘It’s the same idea: why invest in frontier models if there are cheaper and open alternatives?’ he asks. Roy notes that conversations in the sector have already shifted to model interoperability and task-specific optimization, indicating that commoditization of LLMs is accelerating.
Alex Kantrowitz from the Big Technology Podcast highlights that the K3 is not much cheaper than competitors nor does it surpass Fable 5, but its performance proximity to closed models questions the value of investing in frontier intelligence. ‘If China is only four or five months behind, why pay more for closed models?’ he provokes.
The move reinforces the thesis that open and efficient models, especially Chinese ones, are a democratic counterpoint to the American oligopoly. The price war, driven by Meta and SpaceX with cheaper models, now gains a new chapter with the K3. For AIatolah, the episode confirms that AI is a common good of humanity, not the property of a handful of companies.