Archive — US — techcrunch
Cohere claims its new Aya Vision AI model is best-in-class
Cohere has launched its Aya Vision AI model, claiming it's best-in-class for multimodal tasks like image captioning and translation. Available for free via WhatsApp, Aya Vision bridges gaps in AI performance across languages, with models outperforming larger counterparts. Fueled by synthetic annotations for efficient training, the model comes alongside the AyaVisionBench benchmark suite to enhance AI evaluations.
Key points
- Cohere released Aya Vision, a multimodal AI model for tasks like captioning, translating, and summarizing images in 23 languages.
- The model is available for free through WhatsApp, aiming to make AI breakthroughs accessible to researchers.
- Aya Vision includes two versions: Aya Vision 32B and 8B, with 32B outperforming larger models on visual understanding benchmarks.
- Training utilized synthetic annotations to improve efficiency and performance while minimizing resource use.
- Cohere introduced AyaVisionBench to enhance evaluation of AI models, addressing the evaluation crisis in the AI industry.
Why it matters
- Cohere's use of AI-generated synthetic data reflects a broader industry trend due to a shortage of real-world data.
- The AI industry's focus on efficient resource use and multimodal capabilities aligns with emerging IT trends emphasizing inclusivity in AI development.
- Cohere’s efforts can be connected to the 2025 trend of increased accessibility of AI tools for research and practical applications.
Cohere's Aya Vision model represents a significant advancement in AI, especially in its multilingual capabilities and efficient training methods, pushing the boundaries for future AI research and accessibility.