The HiDream-I1-Dev model, developed by the Chinese company Vivago AI, has emerged as the leading open-source image generation model, surpassing FLUX1.1 in the Artificial Analysis Image Arena. HiDream-I1-Dev is now available under the MIT License, allowing for broader access and use. It boasts 17 billion parameters and is designed to produce high-quality images across various styles. The model's open-source release has generated significant interest, with rankings indicating its superiority over other models, including GPT-4o. Additionally, the Kimi-VL model, also recently released, features a lightweight design with approximately 3 billion activated parameters and is noted for its strong multimodal reasoning capabilities, outperforming larger models in specific benchmarks. Both models are accessible on platforms like Hugging Face.
So we got another amazing Vision model that outperforms several sota models like GPT-4o open source and with MIT license?! i take it. https://t.co/RcWvc8iKAP https://t.co/w3eAU0ryDm
A new image Open Source Model is out since Monday and already ranked 3 in Artificial Analysis Image Arena. HiDream-I1-Dev ( there is also two others models Full and Fast ) is an open source model from @vivago_ai . This model open source is also available in @recraftai , where I https://t.co/sDkqxIocLo
Moonshot AI 月之暗面 🌛 @Kimi_Moonshotis just dropped an MoE VLM and an MoE Reasoning VLM on @huggingface 🚀 Model:https://t.co/SS3WKql8xq ✨3B with MIT license ✨Long context windows up to 128K ✨Strong multimodal reasoning (36.8% on MathVision, on par with 10x larger models) https://t.co/dJY7scwaBa