Over the past two years, a Chinese company has been quietly reshaping the landscape of artificial intelligence. While OpenAI, Google, and Anthropic battle to see who charges the most per token, this company... DeepSeek He followed the opposite path: deliver top-tier performance at negligible prices.
The result? In May 2026, DeepSeek launched the V4 Preview — two models (Pro and Flash) that not only rival the best closed-back models in the world, but also cost 10x to 50x less and they are completely open source.
How is DeepSeek so much cheaper?
DeepSeek's history of innovation is a lesson in efficiency. Each generation has brought a leap in cost-effectiveness:
DeepSeek-V2 (2024)
First big impact. It introduced the architecture. MoE (Mixture of Experts) with selective parameter activation. While traditional models trigger 100% of the parameters with each request, DeepSeek activates only the "experts" needed for each task. The result: Performance of a large model at the cost of a small model..
DeepSeek-V3 (2025)
The game-changer. With 671B total parameters (37B active), V3 was trained by $5,5M monthly — compared to hundreds of millions of dollars worth of equivalent models. The market realized: it was no longer acceptable to spend $200 million to train a model when it could be done for 3% of that.
DeepSeek-R1 (2025)
The reality check. While the world applauded OpenAI o1 as a "reasoning revolution," DeepSeek launched R1 with similar chain-of-thought (CoT) capabilities — for a fraction of the costThe announcement sent tech stocks tumbling in the American market. Silicon Valley realized that its monopoly on artificial intelligence was coming to an end.
Technical Innovations That Enable Low Costs
It's not a miracle — it's engineering. DeepSeek has developed three key innovations:
1. Ultra-Efficient MoE Architecture
The V4-Pro has 1,6 trillion total parametersbut it only activates 49 billion by inference — a mere 3%. It's like having 1.600 specialists on the team, but only calling on 49 for each task. The rest remain available, but don't consume resources.
2. Optimized Training
Proprietary parallelization and optimization techniques allow for training giant models with significantly less computing power. The V4-Pro is estimated to have cost... $12M monthly For training. Comparison:
| Córdoba | Training Cost | Difference |
|---|---|---|
| DeepSeek V4-Pro | ~US$ 12 million | |
| GPT-5 | ~US$ 200 million+ | ~17x more expensive |
| Gemini 3 | ~US$ 300 million+ | ~25x more expensive |
3. Open Source = No Artificial Margins
Unlike its American competitors that need to generate returns for investors, DeepSeek operates with much tighter margins. The model is open—anyone can download, inspect, modify, and run it locally. The API exists as a convenience, not as the only option.
The Result: Costs 10x to 50x Lower
In practice, this means that Your company can use cutting-edge AI for pennies.See the comparison of inference costs:
| Córdoba | Cost per 1M tokens (input) | Cost per 1M tokens (output) |
|---|---|---|
| DeepSeek V4-Flash | ~US$ 0,05 | ~US$ 0,15 |
| DeepSeek V4-Pro | ~US$ 0,20 | ~US$ 0,50 |
| GPT-5 | US$ 2,50 | US$ 10,00 |
| Claude 4 (Opus) | US$ 3,00 | US$ 15,00 |
| Gemini 3 Ultra | US$ 1,50 | US$ 5,00 |
V4-Flash costs 50 times less than Claude 4 for generating text. This isn't margin optimization—it's a category change.
And it was precisely this trajectory of innovation that led to the announcement we are about to explore.
🚀 DeepSeek-V4 Preview: The Announcement
On 16 May 2026DeepSeek has released to the public the DeepSeek-V4 PreviewIt's not just another model — it's the consolidation of everything the company has been building: elite performance, extremely low cost, and open source.
The V4 Preview arrives in two versions:
| Specification | V4-Pro | V4-Flash |
|---|---|---|
| Total parameters | 1,6 trillion | 284 bilhões |
| Active parameters | 49 bilhões | 13 bilhões |
| Pre-trained tokens | 33 trillion | 32 trillion |
| Maximum context | 1 million | 1 million |
| Open source | ✅ | ✅ |
| API available | ✅ | ✅ |
| Profile | Expert (precision) | Instant (speed) |
DeepSeek-V4-Pro: Raw Power
With 1,6 trillion parameters With performance that DeepSeek itself claims rivals the best closed-back models in the world, the V4-Pro is designed for tasks that demand the highest level of precision and sophistication.
DeepSeek-V4-Flash: Speed and Economy
With 284B total parameters With only 13 billion active users, Flash is the everyday model. Fast, inexpensive (US$0,05/1M input), and surprisingly capable. Ideal for chatbots, quick analytics, and scalable automation.
1 Million Context Tokens: What Changes
The context of 1 million tokens (approximately 750 words) allows processing in one go:
- ???? Complete books — analyze entire works without dividing them into chapters
- 💻 Entire codebases — understand complete software projects
- 📄 Hundreds of documents Extract insights from PDFs simultaneously.
- ???? Endless conversations Chatbots that remember everything.
- 🏛️ Contracts and manuals — full legal documents
No other open model offers this level of context with consistent quality.
How This Boosts Business
The combination of Low cost + 1M of context + open source It changes the mathematics of several segments:
Automating Customer Service at Scale
Chatbots with 1M token context maintain long conversations without losing history. At $0,05 per 1M tokens, the cost is virtually zero. Senior-quality customer support for pennies.
Legal and Documentary Analysis
Send entire contracts and receive complete analyses: abusive clauses, risks, suggestions. In seconds. Average firms can automate 70% of their document review.
Financial Processing
Quarterly reports, balance sheets, income statements — all processed in batches. Extract key performance indicators, generate executive summaries, and automatically detect anomalies.
Specialized Technical Support
Load manuals, FAQs, and complete documentation into context. Offer top-level technical support without training a model. Great for software, equipment, and healthcare companies.
Corporate Education and Training
Comprehensive learning materials within the context enable AI tutors to master the content from start to finish. Customized training for each employee.
Open Source = Data Sovereignty
For Brazilian companies, the fact that DeepSeek-V4 is open source has an added benefit: You can run the model on your own infrastructure in Brazil.Without relying on external APIs and without sending data outside the country. LGPD-friendly.
Official Links
Conclusion
DeepSeek's trajectory proves that cutting-edge artificial intelligence doesn't have to be expensive. From V2 to V4, the Chinese company innovated in architecture, training efficiency, and business model to deliver... elite performance at affordable prices.
The DeepSeek-V4 Preview is the high point of this journey. With 1 million context tokens, a cost 10x-50x lower than competitors, and open source code, it's not just another release—it's the consolidation of a new era in artificial intelligence.
For Brazilian companies, the message is clear: the technology to transform your business with AI is available, accessible, and within everyone's reach. The question is no longer "if" — it's "how" to do it.
👉 Want to implement generative AI in your business? Talk to Kaizen Agency.
