Archive — US — techcrunch
Google debuts a new Gemini-based text embedding model
Google introduced GeminiEmbedding, a new experimental text embedding model based on its Gemini AI system, designed to enhance text representation in applications like document retrieval. This model outperforms previous versions, supports over 100 languages, and accepts larger text blocks while still being in an experimental phase.
Key points
- Google launched a new embedding model called GeminiEmbedding on its Gemini API.
- Embedding models convert text to numerical representations for semantic understanding.
- GeminiEmbedding is designed for efficiency in applications like document retrieval and classification.
- This model outperforms the previous text-embedding-004 version in various benchmarks.
- GeminiEmbedding supports larger texts and over 100 languages, enhancing its usability.
- Currently, it is in an experimental phase with limited capacity and will evolve before stable release.
Why it matters
- The rise of AI-driven text embedding reflects a broader trend in AI development focused on enhancing language understanding.
- Companies like Amazon, Cohere, and OpenAI are also improving embedding models, indicating a competitive landscape.
- Text embedding models' growing usage aligns with increasing demand for reliable and efficient AI solutions across industries.
- 2025 IT trends emphasize the importance of cloud-based AI models providing scalable and cost-effective solutions.
Google's introduction of GeminiEmbedding highlights its commitment to advancing AI text processing, reflecting ongoing trends in AI development and competition, while promising improved performance in various applications.