Skip to main content

DeepSeek V4 Final Version Leaked, May Launch Next Monday to Take on Kimi K3

Just when everyone thought the AI spotlight was fixed on Kimi K3, DeepSeek sneaks in with a surprise. Test videos and live demos of the final DeepSeek V4 have surfaced online, and whispers suggest an official release as early as next Monday—the first Monday of mid-July. The timing, hot on the heels of K3's hype, has the industry buzzing.

Image

According to leaked test data and feedback from overseas bloggers, the official V4 is no slouch. Early results put it on par with Claude Opus 4.8, especially in 3D generation, where it reportedly leaves Opus in the dust. Agent behavior and coding stability have also seen major upgrades. Some bold claims even suggest V4 outperforms Fable5 and GPT-5.6 in specific tests, like game code generation. One blogger showed off a "Squid Game" mini-game demo: V4 churned out 3,000 lines of code for a grand total of $0.12. That's cost efficiency that'll make you do a double-take.

But it's not just raw power. V4's interaction experience has gotten a subtle facelift. Testers noticed its chain-of-thought style now feels more like Claude—less wall-of-code, more digestible summaries like "writing core functions" and "writing main scene code." It's a readability boost that some analysts attribute to a refined take on Fable's coding style, resulting in cleaner output.

DeepSeek has a track record of working quietly and striking hard on price. If V4 can deliver top-tier performance at a fraction of the cost—orders of magnitude lower—it could spark a "DeepSeek Moment 2.0," echoing the V3 launch. The company had previously hinted at a mid-July release, so next Monday fits the timeline. But as always, we'll wait for the official word.

Key Points

  • DeepSeek V4 final version test videos leaked online, with release possibly next Monday.
  • Performance matches Claude Opus 4.8, with superior 3D generation and coding.
  • Cost efficiency highlighted: 3,000 lines of game code generated for $0.12.
  • Chain-of-thought style improved for readability, resembling Claude's approach.
  • Could trigger another "DeepSeek Moment" if pricing undercuts competitors significantly.