DeepSeek V4 might be released this month, as indicated by the test interface which hints at Vision and Expert modes.

      In the domestic large model sector in China, DeepSeek V4 is clearly one of the most eagerly awaited releases of recent months. Since the launch of its test version in February, users have been anticipating the official debut. Along with a recent rise in system crashes and irregularities, speculation is rife in the market that these may be stress tests leading up to the new version's launch, which could occur as soon as this month. Additionally, a recently leaked screenshot from a gray-scale test has heightened expectations for a significant upgrade.

      A prominent programmer on the Chinese social media site Weibo, @Yigongchang, shared a screenshot of DeepSeek’s updated interface on Tuesday. Unlike the existing version that presents the familiar options of Deep Thinking (R1) and Smart Search (online), the new interface displays three additional modes: Fast, Expert, and Vision. This alteration hints at a redesign of interaction logic and points to a more categorized product lineup. Analysts speculate that the Fast mode may relate to a lightweight model, whereas the Expert mode is likely associated with a deep-reasoning model. The notable introduction of a Vision option further confirms the previously rumored multimodal capabilities, marking a significant advancement for DeepSeek in closing the gap with leading large model rivals.

      In contrast to the current version, the forthcoming DeepSeek release may not consist of a single model but rather a collection of offerings. Based on the interface choices, the new versions could be referred to as DeepSeek V4 Lite (Fast version), DeepSeek V4 (Expert version), and DeepSeek V4 Vision (Vision version). Earlier reports have suggested that DeepSeek is developing at least three large models built on a domestic AI chip-based computing platform, aligning the information from this gray-scale test with previous leaks.

      Although DeepSeek has not yet disclosed specific details regarding the upgrades, the launch of the V4 series seems imminent. Reflecting on last year, the release of DeepSeek R1 made waves in the market due to its high cost-performance ratio, even temporarily influencing NVIDIA's market value.

      Now, with the V4 series achieving multimodal capabilities, domestic expectations are elevated. Some voices in the industry are even advocating for a specialized version focused on AI programming, positioning it to compete directly with Anthropic’s Claude series in coding functionalities. Whether DeepSeek can once again utilize its technological advantages to disrupt the market and apply pressure on Anthropic or OpenAI remains uncertain—answers may be forthcoming in the coming days.

      Jessie Wu is a tech reporter located in Shanghai, covering consumer electronics, semiconductors, and the gaming industry for TechNode. She can be reached via email at jessie.wu@technode.com.