Mistral launches new code embedding model that outperforms OpenAI and Cohere in real-world retrieval tasks
1 min read
Summary
French AI firm Mistral has released its first embedding model, Codestral Embed, which it claims outperforms existing models on benchmarks including SWE-Bench.
The model, which specialises in coding and is available to developers for $0.15 per million tokens, is designed for retrieval-augmented generation (RAG) use cases and to transform code and data into numerical representations.
It can be used for RAG, semantic code search, similarity search and code analytics, according to the company.
“Codestral Embed can output embeddings with different dimensions and precisions,” Mistral said in a blog post.
“The dimensions of our embeddings are ordered by relevance. For any integer target dimension n, you can choose to keep the first n dimensions for a smooth trade-off between quality and cost.