Back to Models

Command R+

Best for RAG
Cohere
Released: 2024-04-04
Type: LLM
Context
128k
Max Out
4.1k
Cutoff
2024-01
Params
104B

The Breakdown

Command R+ by Cohere is the only major model explicitly optimized for RAG (Retrieval Augmented Generation) and tool use as a first-class citizen. Unlike models trained for chat, Command R+ is trained to consume messy documents, extract relevant facts, and generate answers with accurate citations. It drastically reduces hallucinations in enterprise search applications. It's not the model you ask to write a poem; it's the model you ask to summarize a legal contract without getting sued.

Overall Score
8.9
/10
Pricing (per 1k tokens)
Input$0.003
Output$0.015
Currency: USD

The Good

  • Optimized specifically for RAG (Retrieval Augmented Generation)
  • Best-in-class citation and source grounding
  • Strong tool use for enterprise workflows

The Bad

  • Not great at creative writing or 'chatty' personas
  • More expensive than open alternatives like Llama 3
  • Slower than groq-hosted options

The Verdict

The boring, reliable enterprise choice. If you are building a system that reads PDFs and answers questions without making things up, Command R+ is built exactly for that. It cares about citations more than poetry.

Performance Benchmarks

human eval
72
drop
84
mmlu
79.5