Midjourney Announces Video Model Training Plan, V6 Version Coming Soon


Google released the Gemma 4 12B multimodal model, which has 12 billion parameters and innovatively eliminates traditional encoders, allowing direct processing of visual and audio data. This model requires only 16GB of VRAM and can run locally on high-end laptops without relying on cloud resources.
The European Commission recently announced new minimum energy efficiency standards for data centers to address rising energy pressure from surging electricity demand. By 2030, the EU's total data center capacity is expected to increase from 12 GW last year to 28 GW, with electricity consumption exceeding 2.5% of the EU's total power. This move aims to promote energy savings and support sustainable digital services.....
ByteDance open-sources Lance, a native unified multimodal large model with only 3B activated parameters, breaking the technical barriers between understanding models (VLM) and generation models (DiT/Diffusion). It achieves full functionality with extreme lightweight design, challenging the current industry trend of stacking parameters or assembling models, marking an important breakthrough in technological innovation.
Alibaba launches the preview version of Qwen3.6-Max-Preview model, which can be accessed through QwenStudio for conversation or via Alibaba Cloud BaiLian API. Compared to Qwen3.6-Plus, the new model shows significant improvements in agent programming, world knowledge, and instruction following, and performs well in six major programming benchmarks.
In the AI agent field in 2026, the Hermes Agent developed by the Nous Research team has attracted attention, earning over 90,000 stars on GitHub and showing strong market appeal. Compared to the leading open-source agent OpenClaw (Lobster), Hermes is powerful and directly challenges it. OpenClaw is known for its ability to connect various office software, such as Feishu and DingTalk.