Demonstrate intent to support a wide range of enterprise use cases, including cost-effective, resource-intensive big language models (LLMs) — Starting AI Cohere Command has launched the R7B, the smallest and fastest in its R model series.

The R7B command is built to support rapid prototyping and iteration, and uses search-augmented generation (RAG) to improve its accuracy. The model has a context length of 128K and supports 23 languages. It outperformed others in its open-weight model class — Google's Gemma, Meta's Llama, Mistral's Ministral — on tasks like math and coding, Cohere says.

"The model is designed for developers and businesses that need to optimize speed, cost performance, and resource computing for their use cases," Cohere Co-Founder and CEO Aidan Gomes wrote in a blog post announces a new model.

Outranked competitors in Maths, Coding, RAG

Cohere is focused on enterprises and their unique use cases. The company introduced Team R in March and strong R+ command in April, and made updates throughout the year to support speed and efficiency. He teased the Command R7B as the "latest" model in the R series and told the AI ​​research community that the model would weigh in.

According to Koere, an important area in the development of the R7B team is to improve performance in math, reasoning, code and translation. The company seems to have succeeded in these areas, with the new small model taking the lead HuggingFace Open LLM Leaderboard Against similarly sized open weight models such as Gemma 2 9B, Ministral 8B and Llama 3.1 8B.

Furthermore, the smallest model in the R series outperforms competing models in areas such as AI agents, tool usage, and RAG, which helps improve accuracy by grounding model results in external data. Cohere said: Command R7B assistance with negotiation tasks, including technology workplace and enterprise risk management (ERM); technical facts; media workplace and customer service support; HR FAQs; and concluding. Koer also said the model is "exceptionally good" at extracting and manipulating quantitative information in financial settings.

All told, the Command R7B ranked first on average in important metrics, including the post-instruction evaluation (IFeval); big bench hard (BBH); graduate-level Google-proof Q&A (GPQA); multilevel soft thinking (MuSR); and the Understand the language of massive multitasking (MMLU).

Remove unnecessary calling functions

Command R7B can use tools including search engines, APIs and vector databases to extend its functionality. Cohere reports that using the model tool performs strongly against competitors on the Berkeley Function Calling Leaderboard, which evaluates the model's accuracy (connectivity to external data and systems) when calling a function.

Gomez notes that this will prove the model's effectiveness in "real-world, diverse and dynamic environments" and eliminate the need for redundant call functions. This could make it a good choice for building "fast and capable" AI agents. For example, Cohere notes that when operating as an advanced search agent on the Internet, the Command R7B can break down complex queries into smaller targets while performing well in advanced reasoning and information retrieval.

Because it's small, the Command R7B can be used on low-end and consumer processors, GPUs, and MacBooks, allowing for on-device output. The model is now available on the Cohere platform and HuggingFace. The cost is $0.0375 per million input tokens and $0.15 per million output tokens.

"It's an ideal choice for businesses looking for a cost-effective model based on their internal documents and data," Gomez wrote.

Daily insight into business use cases with VB Daily

If you want to impress your boss, VB Daily has got you covered. We give you the inside scoop on what companies are doing with generative AI, from regulatory changes to practical deployments, so you can share insights for maximum ROI.

Read ours Privacy Policy

Thank you for subscribing. See more VB news is here.

An error occurred.