🎯 The Big Picture
Chinese AI lab DeepSeek has unveiled its most ambitious models yet — the V4 series — and they're making waves for both their sheer scale and their radical affordability. With up to 1.6 trillion parameters and prices that undercut every major competitor, DeepSeek is proving the open-weight movement isn't just alive — it's accelerating.
📖 What Happened
DeepSeek launched preview versions of DeepSeek V4 Flash and V4 Pro, the successors to last year's V3.2. Both are mixture-of-experts (MoE) models with massive 1-million-token context windows — enough to ingest entire codebases or lengthy documents in a single prompt.
The flagship V4 Pro boasts 1.6 trillion total parameters with 49 billion active per task, making it the largest open-weight model ever released. That dwarfs Moonshot AI's Kimi K 2.6 (1.1T), MiniMax's M1 (456B), and more than doubles DeepSeek's own V3.2 (671B). The smaller V4 Flash still packs 284 billion parameters (13B active).
DeepSeek claims the new models have "almost closed the gap" with leading frontier models on reasoning benchmarks. The V4-Pro-Max variant reportedly outperforms open-source peers and even beats OpenAI's GPT-5.2 and Gemini 3.0 Pro on select tasks. In coding competitions, both V4 models are said to be "comparable to GPT-5.4."
However, DeepSeek candidly acknowledges the models trail frontier leaders by roughly 3–6 months in knowledge tests — specifically behind GPT-5.4 and Gemini 3.1 Pro. Both models are also text-only, unlike multimodal competitors.
The pricing is perhaps the most disruptive aspect. V4 Flash costs just $0.14 per million input tokens and $0.28 per million output tokens — cheaper than GPT-5.4 Nano, Gemini 3.1 Flash, GPT-5.4 Mini, and Claude Haiku 4.5.
💰 By the Numbers
| 📊 Metric | 💡 Context |
|---|---|
| 1.6T | Total parameters in V4 Pro (largest open-weight model) |
| 49B | Active parameters per task (MoE) |
| 1M | Token context window |
| $0.14 | V4 Flash input price per 1M tokens |
| $0.145 | V4 Pro input price per 1M tokens |
| 3–6 months | Estimated lag behind frontier models |
🎤 Highlights
• V4 Pro is the biggest open-weight model ever released at 1.6T parameters
• Pricing undercuts all frontier competitors including GPT-5.4 Nano and Gemini Flash
• Models show strong reasoning and coding performance but lag in knowledge tests
• Both versions are text-only, no multimodal support yet
• Launch comes one day after U.S. accused China of industrial-scale AI IP theft
🚀 Why It Matters
DeepSeek continues to prove that world-class AI doesn't require Silicon Valley budgets. By releasing massive open-weight models at a fraction of competitors' prices, they're democratizing access to frontier-level capabilities. The 3–6 month gap to state-of-the-art is closing — and for many applications, these models are already "good enough." For developers and enterprises watching their AI bills climb, DeepSeek V4 offers a compelling alternative that could reshape the API economy.
⚡ The Bottom Line
DeepSeek V4 isn't just another model release — it's a price-performance grenade thrown at the frontier AI market. When 1.6 trillion parameters cost less than GPT Nano, the economics of AI get rewritten.
📰 Source: TechCrunch 🔗

