Recently, Moonshot AI officially released its new open-source flagship large model — K3.1Pro. This highly anticipated heavyweight model boasts a parameter scale of up to 2.8T and perfectly supports ultra-long context processing of one million (1M) Tokens, once again refreshing the industry's perceived upper limits for the performance and aesthetic expression of Chinese large models.
Regarding the pace of openness and ecosystem building, Moonshot AI officially confirmed that the complete model weights of K3.1Pro are expected to be fully and officially released before November 4, for developers and researchers worldwide to download and use. Currently, the model has also been launched on OpenRouter, making it convenient for tech enthusiasts and developers to call and test it right away. In authoritative evaluations, K3.1Pro also achieved excellent results on the Artificial Analysis Intelligence Index, ranking fourth on the leaderboard with a score of 55.
In terms of API pricing and version strategy, K3.1Pro demonstrates extremely high cost-effectiveness. The standard version API is priced at 13 yuan per million Tokens for input and 68 yuan for output, nearly halving the cost compared to the previous K3 version. At the same time, officials have also launched a Highspeed version aimed at high-frequency and real-time interaction scenarios. This version achieves an output speed 2 to 3 times that of the standard version, while its price is correspondingly doubled, offering developers with different needs a more flexible and targeted choice.