Underwhelming or underrated? DeepSeek V4 shows “impressive” gains

DeepSeek, a Chinese artificial intelligence firm, has released its long-awaited flagship model, the V4 series, which includes the V4 Flash and V4 Pro. The company touted significant advancements in coding benchmarks and improvements in reasoning and agentic tasks. However, an analysis by benchmark firm Artificial Analysis revealed that the V4 Pro model has not been able to close the gap with its U.S. competitors, ranking second among the leading open-source models globally, behind Moonshot AI’s Kimi K2.6. The V4 series is designed to process longer prompts more efficiently than its predecessor, thanks to a new design that enhances its ability to handle large amounts of text. This feature is a notable improvement over DeepSeek's previous models, which also maintained an open-source approach, allowing broader access to its technology. Despite these advancements, the V4 series has been described as underwhelming in comparison to the market-shaking success of DeepSeek's earlier R1 release. The release of the V4 series comes amid ongoing competition in the AI sector, particularly with U.S. firms leading the market. DeepSeek's struggle to replicate the success of its earlier models highlights the challenges faced by the company in maintaining its position in a rapidly evolving industry. The performance of the V4 Pro, while improved, indicates that DeepSeek still has work to do in order to compete effectively on a global scale.