NewsDeepSeekLarge Language ModelsAI Pricing

DeepSeek V4 Launch Imminent – China's AI Flagship Shocks the Industry

The full version of DeepSeek V4 is expected to go live within days. Early tests suggest Claude-level performance at significantly lower cost.

V4 release expected within days

DeepSeek V4 Launch Imminent – China's AI Flagship Shocks the Industry

DeepSeek V4 is coming. After nearly three months of anticipation, early users confirm that the Chinese AI lab is bringing its next major version into gray testing – with an official release possibly as soon as tomorrow, or within the next few days. Chinese tech portal 36Kr reports based on insider information.

The Essentials

  • Two versions launching: DeepSeek V4 Flash and DeepSeek V4 Pro enter early access
  • First tests show: Performance comparable to Claude Opus 4.8, coding capabilities rival GPT-5.6 Sol
  • Price revolution: DeepSeek introduces "peak-valley pricing" for the first time – significantly cheaper than competitors
  • Recognition trick: Users can test access – the chain-of-thought begins with "I'm" instead of "Let me"

Performance: Strong, but Not Unbeatable

Developer Pankaj Kumar has shared initial test results. The verdict is nuanced: V4 reaches Opus 4.8 level, coding capabilities compete with GPT-5.6 Sol, and agent capabilities are significantly improved. Particularly impressive: 3D and SVG generation work considerably better than before.

But there is criticism too. Some developers report that the Pro version offers no significant advantage over the Flash version. And: on the same tasks, V4 requires more iteration rounds than competitor model Fable 5. Kumar's assessment: V4 will probably not be as powerful as the newly launched Kimi K3 – but significantly cheaper.

The Price Game

Pricing could be the real game-changer. DeepSeek has informed all API users that the V4 launch will also bring a new pricing structure. For the first time, the company introduces peak-valley billing – variable rates depending on load.

Details remain fragmentary, but the pattern is clear: DeepSeek positions itself once again as a price aggressor. While Western providers keep token prices stable or raise them, Chinese competition pursues volume through cost leadership.

The first publicly circulating test demos show impressive results: V4 Pro generated a functional 3D game with siege vehicle mechanics, another demo combines Minecraft and No Man's Sky elements. Classic games like "Cut the Rope" were also created in a single run. SVG rendering and game generation work considerably smoother than in earlier versions.

What This Means for You

The news signals another shift in AI dominance toward Asia. While OpenAI, Anthropic, and Google set benchmarks in the US, DeepSeek systematically builds on cost efficiency and specialized features (3D, SVG, agents). For German companies, this means: the choice between AI providers grows, but price pressure intensifies. Anyone still relying on Western closed-source models should reassess cost dynamics. At the same time, it becomes clear: raw performance alone is no longer enough – price, specialization, and usability become differentiators.

Sources

Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.

Share
← All articles

All analyses are based on i6eal's own measurements or on clearly labelled sources. Figures are snapshots and may change; corrections are disclosed transparently.