Curiosity · AI Model

Command R

Command R, launched by Cohere in March 2024, is the first member of the Command R-series. It targets enterprise retrieval-augmented generation: grounded responses that cite the underlying documents, native tool use, and strong multilingual coverage across 10+ business-critical languages. A 128K context window and permissive per-token pricing made it a popular RAG default.

Model specs

Vendor
Cohere
Family
Command R
Released
2024-03
Context window
128,000 tokens
Modalities
text
Input price
$0.5/M tok
Output price
$1.5/M tok
Pricing as of
2026-04-20

Strengths

  • Strong grounding and citation quality in RAG workflows
  • Native tool use and JSON mode
  • Broad multilingual coverage out of the box

Limitations

  • Raw reasoning below frontier Claude/GPT/Gemini tiers
  • No vision or audio modalities
  • Smaller developer ecosystem than OpenAI/Anthropic

Use cases

  • Enterprise RAG over internal knowledge bases
  • Multilingual customer-support chatbots
  • Tool-using agents on structured business data
  • Grounded question-answering with citations

Benchmarks

BenchmarkScoreAs of
RAG QA (Cohere internal)state-of-the-art for its class at launch2024-03
MMLU≈68%2024-03

Frequently asked questions

What is Command R?

Command R is Cohere's mid-size production LLM, released in March 2024 and optimised for retrieval-augmented generation, multilingual chat, and tool use in enterprise settings.

What makes Command R a RAG-first model?

Cohere trained Command R to cite the supporting documents for its answers, to follow structured tool-use patterns, and to handle long 128K-token contexts — all priorities for enterprise RAG pipelines.

How does Command R compare to Command R+?

Command R+ is the larger, higher-quality sibling. Command R is cheaper and fast; R+ has stronger reasoning and is preferred for harder RAG and agent workflows.

Sources

  1. Cohere — Command R launch — accessed 2026-04-20
  2. Cohere — Pricing — accessed 2026-04-20