Select Language:
After nearly three months of anticipation, the official release of DeepSeek V4 is finally imminent. The full version is expected to drop as soon as tomorrow, or at the latest within the next few days. Already, some users have gained early access to the V4 (General Availability or GA) testing phase, marking a significant milestone for this upcoming AI model.
DeepSeek offers two versions: V4 Flash and V4 Pro. The buzz around these releases is intense, especially following insights from tech insiders and early testers. Notably, the “full-blooded” version of DeepSeek V4 is generating excitement, with online enthusiasts sharing tips on how to identify if they’ve been part of the testing phase. A popular unofficial method involves analyzing the model’s chain-of-thought responses—if the first-person reasoning begins with “I’m” or “I’ll” instead of “Let me,” there’s a good chance the user is interacting with the GA version.
In the current AI landscape, where models like Fable 5 and GPT-5.6 are vying for attention, DeepSeek’s V4 has been receiving notable focus. Developer Pankaj Kumar conducted an early review, comparing V4’s performance to top-tier models. He found that V4 performs comparably to Opus 4.8, trailing slightly behind GPT-5.6 in coding and reasoning tasks, while showing remarkable improvements in 3D rendering and SVG generation. Although V4 requires more iterative cycles for certain tasks than Fable 5, Kumar suggests that amidst fierce competition, V4’s cost-effectiveness might give it an edge, especially considering its likely lower price point compared to newer rivals like Kimi K3.
Meanwhile, the first demo videos of V4’s testing version have begun circulating, drawing diverse opinions. Some see it as capable of rivaling Claude 5, though others note that the Pro edition doesn’t significantly outperform the Flash version. One demo features a 3D shooting game that uses V4 to generate core gameplay mechanics—showcasing the model’s creative potential. Similarly, V4 has produced a hybrid Minecraft and No Man’s Sky-like game experience and even a classic game of “Cut the Rope.” These demos illustrate its rapid development and versatile capabilities, from SVG testing with Xbox controllers to full game creation.
One particularly notable aspect of this launch is the introduction of a “peak-valley” billing system—an innovative pricing model that could fundamentally change how AI services are monetized. DeepSeek announced in late June that, with the release of V4, they would implement dynamic pricing. During peak times, costs per million tokens can double, while off-peak pricing remains significantly lower. For instance, the V4 Pro costs around $0.87 per million output tokens normally, but this can rise to $1.74 during peak periods. Meanwhile, the Flash version’s pricing is even more aggressive, with costs as low as $0.28 per million tokens, making it highly attractive for high-volume users.
This shift toward pay-as-you-go pricing marks a departure from previous flat-rate models, especially impacting enterprise users running continuous agent operations. Still, the low cost of cached responses helps mitigate expenses, as many tasks can be shifted outside of peak hours to achieve better value. Despite this, V4’s remarkable performance at a fraction of the cost of competitors like Fable 5—which prices its 5 million output tokens at $50—underscores DeepSeek’s commitment to offering high value. In controlled benchmarks, V4’s Pro-Max configuration only trails slightly behind Claude Opus 4.6 Max, yet is available at one-seventh the price, reinforcing its appeal.
In addition to the core model, older versions like DeepSeek-Cha t and Reasoner were officially retired on July 24, streamlining their product lineup. Overall, the long wait appears to be nearing its end, and the buzz is that V4 may not claim the absolute top spot in every area, especially against heavyweights like Fable 5 and GPT-5.6 Sol. Still, DeepSeek’s strategy of delivering near-top-tier capabilities at a drastically lower price—leveraging proven performance metrics—is poised to make a big splash, turning what could be an underdog into a game-changer in the AI industry.
ChatGPT
Perplexity
Gemini AI
Grok AI
