Moonshot AI has released Kimi K3, a multimodal model featuring 2.8trn parameters, making it the largest open-weight AI model currently available. Designed for frontier intelligence in coding, knowledge work, and reasoning, K3 delivers performance that competes closely with top American models. According to Artificial Analysis benchmarks, K3 trails behind OpenAI’s GPT-5.6 Sol and Anthropic’s Fable 5, yet it demonstrates superior performance in specific coding and general agent tasks. Furthermore, the model reportedly outperformed GPT-5.6 Sol, GPT-5.5, and Opus-4.8 in comparative assessments.

Kimi K3 maintains a cost-efficient profile at $0.94 per index task and $15 per million output tokens. This pricing sits between Western competitors like GPT-5.6 Sol (max) at $1.04 per task and Claude Fable 5 at $2.75 per task, while remaining significantly higher than Chinese models like DeepSeek’s V4 Pro at $0.04 and MiniMax’s M3 at $0.12. The release follows a $2bn funding round in May, which valued the company at $20bn, backed by investors including Long-Z Investments, Shuimu Capital, China Mobile, and CPE Yuanfeng.

The model launch occurs amid heightening US-China geopolitical tensions regarding AI capabilities and semiconductor export controls. Moonshot AI president Yutong Zhang noted that restricted access to compute resources necessitated a strategic focus on fundamental research and efficiency. Meanwhile, the US continues to implement restrictions on sharing advanced models like GPT-5.6 and Fable, alongside legislative pressure to limit US purchases of Chinese memory chips.

Moonshot AI has released Kimi K3, a multimodal model featuring 2.8trn parameters, making it the largest open-weight AI model currently available. Designed for frontier intelligence in coding, knowledge work, and reasoning, K3 delivers performance that competes closely with top American models. According to Artificial Analysis benchmarks, K3 trails behind OpenAI’s GPT-5.6 Sol and Anthropic’s Fable 5, yet it demonstrates superior performance in specific coding and general agent tasks. Furthermore, the model reportedly outperformed GPT-5.6 Sol, GPT-5.5, and Opus-4.8 in comparative assessments.

Kimi K3 maintains a cost-efficient profile at $0.94 per index task and $15 per million output tokens. This pricing sits between Western competitors like GPT-5.6 Sol (max) at $1.04 per task and Claude Fable 5 at $2.75 per task, while remaining significantly higher than Chinese models like DeepSeek’s V4 Pro at $0.04 and MiniMax’s M3 at $0.12. The release follows a $2bn funding round in May, which valued the company at $20bn, backed by investors including Long-Z Investments, Shuimu Capital, China Mobile, and CPE Yuanfeng.

The model launch occurs amid heightening US-China geopolitical tensions regarding AI capabilities and semiconductor export controls. Moonshot AI president Yutong Zhang noted that restricted access to compute resources necessitated a strategic focus on fundamental research and efficiency. Meanwhile, the US continues to implement restrictions on sharing advanced models like GPT-5.6 and Fable, alongside legislative pressure to limit US purchases of Chinese memory chips.