Artificial intelligence startup Dark Side of the Moon will launch a new model whose capabilities are close to models developed by cutting-edge American artificial intelligence laboratories, such as those developed by Anthropic, showing that the gap between China and the US in the field of cutting-edge artificial intelligence is narrowing. According to two people familiar with the matter, Dark Side of the Moon plans to release Kimi K3 within the next few days. The model's parameter scale will reach 2 trillion to 3 trillion yuan. Anthropic has not disclosed the parameter size of its model, but industry insiders generally speculate that Opus 4.8 has about 1.5 trillion to 2 trillion parameters. People familiar with the matter said that according to mainstream evaluation benchmarks, K3 is expected to surpass Opus 4.8's performance and will be free to download in the form of a so-called “open weight” model, which will also pose a major challenge to American artificial intelligence laboratories such as Anthropic and OpenAI. These companies' high-cost leading-edge models still use a closed source approach.

Artificial intelligence startup Dark Side of the Moon will launch a new model whose capabilities are close to models developed by cutting-edge American artificial intelligence laboratories, such as those developed by Anthropic, showing that the gap between China and the US in the field of cutting-edge artificial intelligence is narrowing. According to two people familiar with the matter, Dark Side of the Moon plans to release Kimi K3 within the next few days. The model's parameter scale will reach 2 trillion to 3 trillion yuan. Anthropic has not disclosed the parameter size of its model, but industry insiders generally speculate that Opus 4.8 has about 1.5 trillion to 2 trillion parameters. People familiar with the matter said that according to mainstream evaluation benchmarks, K3 is expected to surpass Opus 4.8's performance and will be free to download in the form of a so-called “open weight” model, which will also pose a major challenge to American artificial intelligence laboratories such as Anthropic and OpenAI. These companies' high-cost leading-edge models still use a closed source approach.