OpenCode announced that its stealth model, Crouton Alpha, will be free for the next week. The model features a 1M context window and multimodal capabilities, with a promise not to train on user data, though access is limited to specific regions.
Multimodal
多模态
Models capable of processing and understanding multiple types of data inputs simultaneously.
Curated Intelligence
Meta has released Muse Spark Open, an 816-billion-parameter multimodal model with a 1M-token context window, claiming top performance among US open-weight models. Designed for coding and agentic tasks, it natively supports Apple Silicon but requires at least 1TB of RAM for local execution.
Meta has released Muse Spark Open, an 816-billion-parameter open-weight multimodal model capable of processing text, images, video, and audio with a 1-million-token context window. It performs on par with leading Chinese open-source models like Qwen and GLM in benchmarks and is specifically optimized for local execution on Apple's M5 Ultra chip.
Jev 1.14 is now available on OpenRouter, introducing experimental multimodal vision capabilities and significant multilingual improvements. The release maintains its low-cost pricing structure while delivering incremental performance gains across the board.
Meta has released Muse Spark Open, the most capable US open-weights multimodal model to date. Featuring a Mixture-of-Experts architecture with ~816B total parameters (~42B active per token) and a 1M-token context window, it processes text, images, audio, and video while being specifically trained for agentic coding tasks.
ByteDance's Doubao Work platform introduced an infinite canvas feature for organizing assets and iterative design editing. It also integrated the new Seedream 5.0 Flash image model and launched the Doubao 2.1 Lite model to enhance multimodal creation efficiency in high-frequency scenarios like document and PPT generation.
Google released AI Edge Foresight, a macOS app powered by the on-device multimodal model EmbeddingGemma 2. The tool operates entirely offline to generate polished meeting minutes from audio and notes, while also supporting real-time Q&A based on meeting content.
Google released the open-source multimodal embedding model EmbeddingGemma 2, which outperforms rivals twice its size on benchmarks despite having only 740 million parameters. The model supports local execution with minimal RAM usage (191 MB), significantly reducing vector database storage requirements and improving retrieval efficiency.
Microsoft Research introduces Quine, an AI system building a multimodal world model of biology spanning genomics, proteins, and cell states. The system creates a closed loop between in silico computation and wet-lab experiments, successfully prioritizing and validating therapeutic compounds for tumor-state shifts in collaboration with the Broad Institute.
Alibaba's Qwen platform has launched 'Qwen AI Arena', a challenge and evaluation platform designed to test AI Agents in real-world business scenarios. The inaugural task focuses on cross-border e-commerce product listing, requiring developers to build Agents capable of generating multilingual copy and multimedia assets simultaneously, supported by provided tokens and multimodal models.