French AI startup Mistral has unveiled Pixtral Large, a new 124-billion-parameter language model, alongside major updates to its Le Chat platform, reports Carl Franzen. The new model features advanced multimodal capabilities, including image processing and optical character recognition, while maintaining a significant context window of 128,000 tokens. The model is available for research purposes through …
Moondream, a startup backed by Felicis Ventures, Microsoft’s M12 GitHub Fund, and Ascend, has emerged from stealth with $4.5 million in pre-seed funding. According to VentureBeat’s Michael Nuñez, the company has developed an open-source vision-language model that boasts 1.6 billion parameters but matches the performance of models four times its size. The model, which can …
Meta has launched Spirit LM, its first Spirit LM is Meta’s first freely available multimodal language model that integrates text and speech inputs and outputs, positioning it as a competitor to models like OpenAI’s GPT-4o. Developed by Meta’s Fundamental AI Research (FAIR) team, Spirit LM aims to enhance AI voice experiences by improving speech generation’s …
ARIA is an open multimodal native mixture-of-experts model designed to integrate diverse forms of information for comprehensive understanding, outperforming existing proprietary models in various tasks. With 24.9 billion total parameters, it activates 3.9 billion and 3.5 billion parameters for visual and text tokens, respectively. The model is pre-trained on a substantial dataset comprising 6.4 trillion …
Nvidia has released a powerful open-source AI model that rivals proprietary systems from industry leaders like OpenAI and Google. The model, called NVLM 1.0, demonstrates exceptional performance in vision and language tasks while also enhancing text-only capabilities. Michael Nuñez reports on this development for VentureBeat. The main model, NVLM-D-72B, with 72 billion parameters, can process …
Meta has today released the new version of its AI model series: Llama 3.2, which for the first time includes vision models that can process both images and text. The larger versions with 11 and 90 billion parameters should be able to compete with closed systems like Claude 3 Haiku for image processing. Also new …
French AI startup Mistral has released its first multimodal model, Pixtral 12B. In other words, it has 12 billion parameters and can process both images and text. It is based on Mistral’s existing text model Nemo 12B and is said to be able to answer questions about any number of images of any size. Pixtral …
The new “Multimodal Arena” from LMSYS compares the performance of different AI models on image-related tasks and shows that OpenAI’s GPT-4o leads the pack, closely followed by Claude 3.5 Sonnet and Gemini 1.5 Pro. Surprisingly, open source models such as LLaVA-v1.6-34B achieve results comparable to some proprietary models. The catch? Despite progress, Princeton’s CharXiv benchmark …
The “4M” AI model provides a glimpse into Apple’s progress in artificial intelligence. Developed in collaboration with EPF Lausanne, the model can convert text to images, recognize objects, and manipulate 3D scenes based on speech input.
Facebook’s parent company Meta has unveiled Chameleon, a new multimodal AI model that can process images, text, and code simultaneously. Unlike other models that use separate components for different types of data, Chameleon was designed from the ground up to handle multiple modalities.