Curiosity · AI Model

Command R+

Command R+ is Cohere's 104B open-weights flagship — released in multiple refreshes through 2024–2025 and purpose-engineered for retrieval-augmented generation, tool use, and enterprise workflows. Weights are public for research under CC-BY-NC, with commercial production served through Cohere's API.

Model specs

Vendor
Cohere
Family
Command R
Released
2024-08
Context window
128,000 tokens
Modalities
text
Input price
$2.5/M tok
Output price
$10/M tok
Pricing as of
2026-04-20

Strengths

  • Best-in-class grounded citation behavior for RAG
  • Strong structured function calling and tool use
  • Broad multilingual coverage — English, French, Spanish, Italian, German, Portuguese, Japanese, Korean, Arabic, Chinese
  • Open weights available on Hugging Face for research and evaluation

Limitations

  • CC-BY-NC license blocks commercial self-host — must use Cohere API for production
  • Trails newer open models on general reasoning benchmarks
  • Per-token API pricing higher than open self-host alternatives
  • Text-only — no native vision or audio

Use cases

  • Enterprise RAG pipelines with high-fidelity grounded citations
  • Multi-step tool-use agents with structured function calling
  • Multilingual customer support across ten major languages
  • Document-heavy workflows with 128K context

Benchmarks

BenchmarkScoreAs of
MMLU≈75%2024-08
HumanEval≈70%2024-08
MIRACL (RAG)≈62%2024-08

Frequently asked questions

What makes Command R+ different?

Command R+ is tuned specifically for enterprise RAG — grounded citations, structured tool use, and multilingual retrieval quality. It's engineered to be a drop-in generation engine for vector-DB pipelines rather than a general chat model.

Is Command R+ truly open-source?

Weights are published on Hugging Face under CC-BY-NC — free for research and non-commercial use. Commercial production deployments require Cohere's paid API, so it's 'open weights' but not open-source in the OSI sense.

When should I pick Command R+ over Llama 3.3 70B?

Pick Command R+ when RAG fidelity and built-in citation support are the main requirement, or when you want Cohere's tool-use ergonomics. Pick Llama 3.3 70B for fully open self-host and broader community support.

Sources

  1. Cohere — Command R+ blog — accessed 2026-04-20
  2. Hugging Face — CohereForAI/c4ai-command-r-plus — accessed 2026-04-20