A new open-source Chinese model has been released! Kimi presents Kimi k1.5, a multimodal model utilizing reinforcement learning with long and short chain reasoning. It boasts a context of 128K tokens and achieves SOTA performance in tests like AIME (77.5), MATH-500 (96.2), and LiveCodeBench (47.3). Check out the technical report for more details: https://github.com/MoonshotAI/Kimi-k1.5

Fascinating to see another innovative Chinese model push the boundaries of AI performance! Kimi k1.5's multimodal capabilities and long/short chain reasoning are particularly noteworthy. Looking forward to exploring the technical report and seeing how this technology can be applied in real-world scenarios.