project
Command A+ - Cohere Labs' open-source multimodal large model
Command A+ is an open-source multimodal large model from Cohere Labs. It employs a hybrid expert architecture, has a total of 218 bytes of parameters but only 25 bytes of activation parameters, balancing powerful performance with inference efficiency. The model supports 128K ultra-long sequences...
What is Command A+?
Command A+ is an open-source multimodal large model from Cohere Labs. It employs a hybrid expert architecture with a total of 218 bytes of parameters and only 25 bytes of activation parameters, balancing powerful performance with inference efficiency. The model supports 128K bytes of context and 48 languages, natively integrating visual understanding, agent task execution, complex inference, and machine translation capabilities. It offers three quantization levels, requiring only one B200 or two H100 quantization sheets for private deployment. The model is freely commercially viable and can be further developed, making it an ideal foundation model for enterprise-level agent applications, retrieval enhancement generation, and multimodal AI applications.
The main functions of Command A+
- Multimodal AgentIt natively supports tool calls and multi-step Agent tasks, and performs best in tool usage and autonomous execution.
- Visual inputIt supports combined image and text input to achieve image and text understanding and generation.
- Reasoning and TranslationBuilt-in inference chain and world-class translation capabilities, covering 48 languages including Chinese, English, and French.
- Long context processing128K input context and 64K output length, suitable for long document RAGs and complex workflows.
- Efficient deploymentPrivate deployment requires only 1×B200 or 2×H100 GPUs, with 30% lower latency than Command A Reasoning.
The technical principles of Command A+
- MoE architectureIt adopts a hybrid expert architecture with a total of 218B parameters and only 25B activation parameters, balancing performance and inference efficiency.
- Unified weightIt integrates vision, reasoning, translation, and agent capabilities into a single model weight, eliminating the need for multiple models to be combined.
- RAG native optimizationIt forms an integrated retrieval enhancement generation stack with Cohere Embed v3 and Rerank 3.5, supports native citation tracing, and has a significantly lower illusion rate than general models.
How to use Command A+
-
Download weightUse the Hugging Face CLI tool to pull BF16, FP8, or W4A4 quantized version model files to your local machine.
-
Transformers loadingUse the pipeline interface of the Hugging Face Transformers library to directly load the model for graph-text inference.
-
vLLM DeploymentAfter installing vLLM, specify the model path to start the service, which will automatically expose API endpoints in an OpenAI-compatible format.
-
SGLang DeploymentAfter installing SGLang, start the inference service, which also provides an OpenAI-compatible API for external calls.
-
Docker runningExecute Docker commands to start containerized services with a single click, without the need to manually configure dependencies.
-
API callsSend a standard Chat Completions request to the local service, and passing text and image URLs in the message body can trigger inference.
Command A+'s core advantages
-
Enterprise-level efficiency: Throughput is increased by 110% and latency is reduced by 30% compared to the previous generation, designed for large-scale production deployment.
-
Extremely low deployment thresholdW4A4 quantization only requires 1 B200 or 2 H100 for private deployment.
-
Data sovereigntySupports local deployment and mainstream cloud platforms, meeting compliance requirements in finance, healthcare, and other sectors.
-
Open source friendlyApache 2.0 License, with full open authority, code, and tuning tools; unrestricted for commercial use.
Command A+'s project address
- Project official websitehttps://cohere.com/blog/command-a-plus
- HuggingFace model libraryhttps://huggingface.co/CohereLabs/command-a-plus-05-2026-bf16
Comparison of Command A+ with similar competitors
| Dimension | Command A+ | OpenAI GPT-4o |
|---|---|---|
| Developer | Cohere Labs | OpenAI |
| Open source license | Apache 2.0(Fully Commercial) | Closed-source / proprietary API |
| Model Architecture | MoE (218B / 25B) | Dense (not publicly disclosed) |
| Multimodal capabilities | Visual + Text + Tools | Visual + Audio + Text |
| Context window | 128K Input / 64K Output | 128K |
| Private Deployment | 1×B200 or 2×H100 | API calls only |
| API Price (per million tokens) | $2.50 / $10.00 | $2.50 / $10.00 |
| Native RAG ecosystem | (Embed + Rerank + Generate) | Self-built required |
| Native Agent tool call | ||
| Enterprise-grade RAG stack | end to end | |
| Quantitative support | BF16 / FP8 / W4A4 | not applicable |
Application scenarios of Command A+
-
Enterprise Knowledge Base Q&ABased on a 128K long context and native RAG capabilities, it enables intelligent question answering across documents and includes precise citation tracing.
-
Multilingual translation and localizationIt covers 48 languages and is suitable for cross-border e-commerce, international conference simultaneous interpretation, and multilingual content production.
-
Visual Agent AutomationBy combining image understanding and tool usage, it enables end-to-end automated workflows for invoice recognition, invoice entry, and report parsing.
-
Compliant industry privatization deploymentFinancial, healthcare, and government institutions can deploy on-premises or in private clouds to meet data sovereignty and regulatory compliance requirements.
-
High-throughput intelligent customer serviceLeveraging the high throughput and low latency characteristics of the MoE architecture, it supports large-scale concurrent dialogues and real-time retrieval enhanced responses.