Open-source LLM · Cohere
Command R+
104 B parameters · 128K context · CC-BY-NC-4.0 (non-commercial) · released 2024-04-04 by Cohere.
Command R+ needs about 208 GB at FP16: no single GPU can hold it. Serving requires an 8-GPU HGX baseboard (INT8/INT4) or a full rack (NVL72, TPU pod) at FP16.
Where this model sits
Built for RAG and tool use with strong multilingual coverage (23 languages). Non-commercial license is the big constraint on adoption.
What runs it (production serving)
Datacenter GPUs and rack designs sized for Command R+. Consumer cards are not practical at this size.
Smallest datacenter GPU that fits (INT4)
Multi-GPU / rack-scale (FP16)
Memory at different context lengths
Weights are fixed. The KV cache grows linearly with context length and can rival the weights at full window. Most guides only quote weights, then hand-wave a 15% overhead, which is only true at short context.
| Context | FP16 total | INT8 total | INT4 total |
|---|---|---|---|
| 4K (short chat) | 208 + 1.0 = 209.0 GB | 104 + 0.5 = 104.5 GB | 52 + 0.5 = 52.5 GB |
| 32K (typical RAG) | 208 + 8.0 = 216.0 GB | 104 + 4.0 = 108.0 GB | 52 + 4.0 = 56.0 GB |
| 128K (full window) | 208 + 32.0 = 240.0 GB | 104 + 16.0 = 120.0 GB | 52 + 16.0 = 68.0 GB |
KV bytes per token = 2 × 64 layers × 8 KV heads × 128 head-dim × precision bytes. INT4 rows keep the KV cache at INT8 (typical serving default).
Quality benchmarks
Vendor-reported scores from the release blog or tech report. Every row cites its source.
| Benchmark | Score | Source | Date |
|---|---|---|---|
| MMLU (5-shot) | 75.7 | Cohere Command R+ blog | 2024-04-04 |
| HumanEval | 70.7 | Cohere Command R+ blog | 2024-04-04 |
Speed and cost to serve
What cloud providers currently charge to serve Command R+. Tokens per second is what the provider publishes on their own page. Batch size, prompt length, and inference engine can shift these numbers a lot; treat as directional.
| Provider | Hardware | Speed | $/M in | $/M out | As of |
|---|---|---|---|---|---|
| Cohere API | — | — | $2.50 | $10.00 | 2025-08-01 |
| AWS Bedrock | — | — | $3.00 | $15.00 | 2025-08-01 |
Cheapest option in this list for output-heavy workloads: Cohere API at $10.00/M out. Fastest published: not enough data.
License and variants
- License
- CC-BY-NC-4.0 (non-commercial)
- The gotcha
- Non-commercial only, so you cannot serve this to paying users without a separate commercial license from Cohere.
- Variants
- Instruct
- Context window
- 131,072 tokens (128K)
- Architecture
- dense
Compare with
Model card: https://huggingface.co/CohereForAI/c4ai-command-r-plus · Release: Cohere announcement
See also: every open-source LLM we track · all AI chips · rack designs.