Open-Source AI Models Rival Closed Frontier

Kimi K3 (Moonshot AI, 2.8T parameters) matches GPT-5.6 Sol on benchmarks and leads on agent tasks; Alibaba's Qwen3.8-Max (2.4T, open-weight) competes with Anthropic's Fable 5 and includes QwenWork enterprise agent launch; Tencent's Hy3 (open-weight) leads on GPQA and SWE-Bench. These releases shift enterprise build-vs-buy calculus, offering privacy and cost advantages. Mistral also released Shieldstral, a 3B open-weights multimodal moderation model.