DeepSeek V4's official version was delayed after its gradual testing started in mid-July, and the official has not yet announced the exact release date. However, multiple new clues point to August 3rd being a key node - the Silicon Valley Flow platform first exposed the secret.
The platform announced that it will raise the cache hit price of DeepSeek V4 Pro model from the current 0.1 yuan per million Tokens to 1.0 yuan starting on August 3rd. Such an abrupt price increase can only be reasonably explained by the imminent release of the official version of DeepSeek V4; otherwise, it is hard to explain the motivation behind the Silicon Valley Flow's actions.
APIs become 'smarter' ahead of time, frequent service fluctuations confirm the gradual testing
Another key signal comes from the developer community's test verification. Several netizens found that some DeepSeek V4-related APIs can now correctly answer the release dates of models such as Claude Opus 4.5 and Sonnet 4.5, which were previously unable to be answered without internet access. The tests were conducted through APIs rather than web chats, and not all users could reproduce them, which is a typical feature of gradual testing with phased releases.
Recently, DeepSeek services have frequently experienced instability, which also indirectly confirms that the team is continuously conducting small-scale tests and rolling back changes to refine stability. Since mid-July, DeepSeek has been continuously iterating, and the external expectation is that the official version's performance should at least reach the level between GLM 5.2 and Claude Opus 4.8.
Competitors are not idle: OpenAI cuts prices by 80%, GLM increases prices
The pressure on DeepSeek V4 is not limited to the technical side. OpenAI announced today that the GPT-5.6 series has been reduced by up to 80%, and the cost-effectiveness after the price cut is almost comparable to DeepSeek V4 Pro, combined with multi-modal advantages, the title of the most cost-effective model has already been taken by OpenAI. In contrast, the latest subscription price of GLM has increased by 2 to 3 times in China, making the cost-effectiveness advantage of GPT even more prominent. Under this differentiated situation where competitors are cutting prices to capture the market and raising prices to increase revenue, whether DeepSeek V4 can regain ground with performance may be answered on August 3rd.
