Cohere has released its most powerful AI model yet as an open-source offering, making it freely available for developers and enterprises. The new model, named Command R+, is designed to outperform previous versions in reasoning, long-context tasks, and multilingual capabilities, with a focus on enterprise-grade retrieval-augmented generation (RAG). The release marks a significant shift for Cohere, which had previously kept its strongest models proprietary.
Command R+ is available under a permissive license on platforms like Hugging Face and GitHub, allowing commercial use with some restrictions. The model is optimized for factual accuracy and reduced hallucination, particularly in knowledge-intensive tasks.
What Command R+ Delivers
Cohere claims Command R+ achieves leading results on benchmarks for RAG and tool use. Key features include:
- Long-context window of up to 128,000 tokens, enabling processing of entire documents without truncation.
- Multilingual support across 10 languages, including English, Spanish, French, and Japanese.
- Reduced hallucination through a technique called “grounded generation” that forces the model to cite sources in its outputs.
The model is designed to be deployed on standard GPU hardware, lowering the barrier for organizations that cannot afford massive compute clusters.
Why Open-Sourcing Matters
Cohere’s decision to open-source its strongest model to date contrasts with rivals like OpenAI and Anthropic, which keep their most advanced systems closed. The company argues that transparency and community collaboration are essential for safe and responsible AI development.
“By open-sourcing Command R+, we are giving the developer community the ability to inspect, improve, and build upon the most capable model we have ever created.”
The move also positions Cohere to compete with other open-source leaders like Meta’s Llama 3 and Mistral AI, both of which have released high-performance open models.
Performance and Benchmarks
In internal evaluations, Command R+ scored higher than Cohere’s previous flagship, Command R, on tasks like summarization, question answering, and code generation. On the MMLU benchmark, which measures general knowledge, it surpassed several closed-source models of similar size.
However, the model still lags behind the largest proprietary models from Google and OpenAI on certain complex reasoning tests. Cohere acknowledges this and frames Command R+ as a practical, deployable alternative rather than a record-breaker.
Licensing and Restrictions
The open-source license allows modification and redistribution, but it prohibits use in safety-critical applications without additional safeguards. Cohere also requires users to adhere to its acceptable use policy, which bans malicious or deceptive applications.
This middle-ground approach aims to encourage innovation while limiting potential misuse. Developers can freely integrate the model into commercial products, but must take responsibility for ensuring ethical deployment.
Enterprise and Developer Impact
For enterprises, Command R+ offers a cost-effective way to build AI-powered search, customer support, and content generation tools without vendor lock-in. The model’s focus on RAG means it can be paired with private databases to produce grounded, verifiable answers.
Developers gain access to a state-of-the-art model that runs on consumer-grade GPUs with quantization, making local experimentation feasible. The open weights also allow fine-tuning on proprietary data, a major advantage over API-only models.
Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge.
What are your thoughts on this? I’d love to hear about your own experiences in the comments below.