**[DeepSeek V4 "Full-Power Edition" Revealed, Set to Launch as Early as Tomorrow]** According to a report by Jinse Finance, market sources indicate that the official version of DeepSeek V4 could be released as early as tomorrow. Currently, some individuals have already gained early access to the DeepSeek V4 (GA) beta testing. There are two versions available: DeepSeek V4 Flash and DeepSeek V4 Pro. Blogger AiBattle shared a popular "inspection rhyme" for identifying the versions: observe the first-person phrasing in the Chain of Thought (CoT). If the model's thought process begins with "I'm" or "I'll," rather than the older version's "Let me," it indicates the new version. Its performance is comparable to Opus 4.8, with encoding capabilities rivaling GPT-5.6Sol, but at a lower price. The API introduces peak-valley pricing, with peak rates doubling and cache hit rates being extremely low. While its performance may not rank first across all metrics, it remains a price disruptor, reducing Opus-level capabilities to one-seventh of the cost. (Source: New Intelligence Yuan)