Recently, Cohere announced a comprehensive upgrade to its Command R and Command R+ AI models, significantly enhancing their capabilities in enterprise applications. The new versions, command-r-08-2024 and command-r-plus-08-2024, have made significant progress in retrieval-augmented generation (RAG), multilingual support, and adaptive tool usage.

In this update, the standard version of the Command R model has approached or even surpassed its predecessor, Command R+. The new model achieves a 50% increase in throughput and a 20% reduction in latency, while halving the hardware requirements, bringing more efficient and cost-effective AI operations solutions to enterprises.

The new series of models support up to 23 languages and perform exceptionally well in multilingual RAG, which is particularly important for global enterprises. By providing accurate cross-language references, misleading information in AI-generated content is significantly reduced, improving the accuracy and reliability of the content.

Cohere has also optimized the models' capabilities in programming, mathematics, and logical reasoning, particularly in the significant improvements in structured data analysis. It can delve deeper into various data formats, including tabular data, providing strong support for data-driven decision-making in industries such as finance, SaaS, and consulting.

To enhance the flexibility of AI governance, Cohere has introduced a beta version of Safety Modes, allowing enterprises to switch between "strict" and "contextual" security settings to adapt to different risk tolerances and application scenarios.

In addition, the new models introduce multiple features to improve usability and customization options, including improved structured output to enhance the accuracy of JSON generation and enhanced compliance with instructions. At the same time, the models' sensitivity to non-semantic prompt changes has been reduced, ensuring more stable output results.

Currently, these upgraded models are available through Cohere's hosted API and Amazon SageMaker at competitive prices, and will be expanded to more cloud platforms in the future. The specific pricing is as follows: command-r-08-2024 $0.15 per million input tokens, $0.60 per million output tokens; command-r-plus-08-2024 $2.50 per million input tokens, $10.00 per million output tokens.