Cohere Releases Command A+: Open-Source MoE Model for Enterprise AI
Key Takeaways
- ▸Command A+ is now available open-source under Apache 2.0, enabling developers to run, control, and adapt an enterprise-grade model locally
- ▸The MoE model achieves significant performance gains over previous Command A versions: 48% improvement on telecom reasoning, 8x improvement on agentic coding, and substantial gains on multimodal and multilingual tasks
- ▸Designed for efficiency with support for quantization and lightweight inference, addressing cost and deployment constraints in enterprise AI workflows
Summary
Cohere has released Command A+, an open-source mixture-of-experts (MoE) language model designed for high-performance agentic tasks with minimal compute overhead. Available under an Apache 2.0 license on Hugging Face, the model represents the consolidation of capabilities from across Cohere's Command series, refined through a year of real-world deployment via the North enterprise workspace.
Command A+ significantly outperforms previous Command A generations across critical benchmarks. The model achieved a 𝜏²-Bench Telecom score improvement from 37% to 85%, terminal-bench hard coding performance of 25% (up from 3%), and 20-32% improvements on agentic question answering and spreadsheet analysis tasks. Additionally, the model nearly doubles performance on multilingual tasks, expanding language coverage from 23 to 48 languages.
With optimizations for low-bit quantization, efficient inference, and integration with open inference frameworks, Command A+ is positioned as a developer-friendly option for on-premise and sovereign AI deployment. The release underscores Cohere's commitment to making enterprise-grade agentic capabilities accessible without proprietary infrastructure dependencies.
- Represents Cohere's vision of 'sovereign AI'—democratizing access to advanced agentic capabilities beyond proprietary platforms
Editorial Opinion
Command A+'s open-source release is a meaningful step toward AI independence in the enterprise market. By offering an efficient, agentic-focused model under a permissive license, Cohere is directly challenging the closed-model ecosystem while addressing a real pain point: the ability to deploy and control advanced AI capabilities on-premise. The performance gains in agentic tasks (especially coding) suggest this model is genuinely useful rather than a mere open-source gesture—critical for adoption.
