Blog

The battle of AI dominance continues to heat up as new models enter the scene, and this past week, DeepSeek made headlines for its promising performance as an open-source alternative. Our latest LLM Evaluation Report (The Lord of the LLMs) systematically assessed top AI models, including OpenAI’s GPT, Google’s Gemini, Anthropic’s Claude, Meta’s Llama, and several others, across five key domains: Writing, Coding, Retrieval-Augmented Generation (RAG), Search, and Graphics. While GPT led the pack, DeepSeek emerged as an interesting contender, demonstrating solid writing and coding capabilities despite being an open-source model. However, its responses bore an uncanny resemblance to GPT—even when it made mistakes.

DeepSeek’s Writing (87), Coding (85), and RAG (90) scores highlight its competency in handling text-based tasks. Compared to other open-source models like Llama and Mistral, it delivered strong performance and proved to be a viable option for those seeking an alternative to proprietary AI systems. That said, one of the more curious findings was how often DeepSeek’s answers mirrored GPT—even down to replicating the same errors. This raises questions about its training data and whether it was heavily influenced by OpenAI’s outputs, either directly or indirectly.

While DeepSeek holds promise, our evaluation suggests that it still lacks distinctiveness and independent reasoning compared to leading AI models. As businesses and developers consider integrating LLMs into their workflows, they must weigh the advantages of open-source flexibility against the potential limitations in originality and accuracy. For a deeper look at how DeepSeek and other top AI models performed in our rigorous testing, check out the full LLM Evaluation Report here.