Zhipu GLM-5.1 en 2026 : Le modèle ouvert entraîné sans un seul GPU Nvidia
GLM-5.1 is the Claude alternative that costs $3 a month and was trained without any Nvidia hardware. Zhipu’s model is a geopolitical statement as much as a technical one.
L'IA chinoise
Moonshot Kimi K2.6 en 2026 : le modèle ouvert qui surpasse le GPT-5.5
Moonshot’s Kimi K2.6 did something no open model had done before: beat a frontier US model at real-world coding. Here’s how the Beijing startup became the agent-coding leader.
DeepSeek en 2026 : comment un laboratoire chinois est devenu le roi de l'IA en termes de prix et de performances
DeepSeek went from unknown quant-fund side project to the model that reset global AI pricing. Here’s what DeepSeek V4 actually is in 2026, and the honest case for and against using it.
ByteDance Doubao en 2026 : l'application d'IA la plus utilisée en Chine, expliquée
ByteDance turned its TikTok-scale distribution into the world’s 4th-largest AI app. Doubao isn’t chasing benchmarks — it’s chasing hundreds of millions of users. Here’s how.
Alibaba Qwen en 2026 : la famille de modèles d'IA la plus complète au monde
Alibaba’s Qwen is the quiet giant of Chinese AI — the broadest model family in existence, an open-weight ecosystem rivaling Meta’s, and a flagship that just cracked the global top 10.