Command R+
Cohere's RAG- and tool-use-optimized model, still live but superseded by Command A.
Context window
128K tokens
Input / 1M tokens
$2.50
Output / 1M tokens
$10.00
Provider
Cohere
$2.50 per 1M input tokens and $10.00 per 1M output tokens (Command R+ 08-2024). Cohere bills only for tokens processed, not for context length. · Data verified 2026-07-02
Command R+ (08-2024 build) is Cohere's enterprise model optimized for retrieval-augmented generation (RAG) and multi-step tool use. Released August 2024 with a 128K-token context window and 4K max output, it grounds responses in supplied documents with citations, sequences tool calls for agentic workflows, and supports 10 optimized languages. The 08-2024 update delivered roughly 50% higher throughput and 25% lower latency over the prior version. It remains available but Cohere now points most users to its newer flagship, Command A (command-a-03-2025).
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Call the Cohere API (Chat endpoint) with model 'command-r-plus-08-2024' using the official 'cohere' SDK, or access via Amazon Bedrock. Note Cohere now recommends its newer flagship 'command-a-03-2025' (Command A) for most use cases.
Strengths
- ✓Best-in-class retrieval-augmented generation with inline citations
- ✓Robust multi-step tool use for agentic workflows
- ✓128K-token context window
- ✓Multilingual support optimized for 10 languages
- ✓Available via both Cohere API and Amazon Bedrock
Best for developers who...
When to choose it (and when not to)
Reach for Command R+ when...
- →Your core use case is grounded RAG with citations
- →You need reliable multi-step tool/agent orchestration
- →You want an enterprise model available on Bedrock
- →You need multilingual responses across the 10 optimized languages
Look elsewhere if...
- ✕You want Cohere's newest and most capable model (choose Command A)
- ✕You need outputs longer than 4K tokens
- ✕You want multimodal (image/audio) input
- ✕You need the lowest possible cost per token
How to use it
- ›Pass source documents through the Chat endpoint's documents field to get grounded, cited answers
- ›Define clear tool schemas to enable reliable multi-step tool use
- ›Rely on the model's ability to decline unanswerable questions to reduce hallucination
- ›For non-English tasks, prompt in one of the 10 optimized languages
Quickstart
Pythonimport cohere
co = cohere.ClientV2(api_key="YOUR_COHERE_API_KEY")
resp = co.chat(
model="command-r-plus-08-2024",
messages=[{"role": "user", "content": "Draft a short product update email."}],
)
print(resp.message.content[0].text)Requires the official 'cohere' Python SDK (pip install cohere). Command R+ is also available on Amazon Bedrock as cohere.command-r-plus-v1:0.
API model id: command-r-plus-08-2024
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| RAG (BEIR) | Top-5 | Competitive on standard RAG benchmarks |
| MMLU | ~75% | Not the primary benchmark focus |
Source: Cohere Command R+ documentation
Compare Command R+ with
Command R+ vs Claude Sonnet 4.6
Anthropic - 1M ctx
Command R+ vs Mistral Large
Mistral AI - 128000 ctx
Command R+ vs Amazon Nova Pro
Amazon Web Services - 300K tokens ctx