Three reasons why DeepSeek’s new model matters

2026-04-24 · Source: MIT Technology Review · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Emerging Technologies & Innovation · Depth: Intermediate, medium

Summary

Chinese AI firm DeepSeek has released a preview of V4, its new flagship open-source model, which significantly improves long prompt processing efficiency through architectural changes to its attention mechanism. V4 comes in two versions: V4-Pro, designed for coding and complex agent tasks, and V4-Flash, a faster, cheaper alternative. Both versions offer reasoning modes and boast competitive performance against leading closed-source models like Anthropic's Claude-Opus-4.6, OpenAI's GPT-5.4, and Google's Gemini-3.1 on major benchmarks, while outperforming other open-source models. DeepSeek V4-Pro is priced at $1.74 per million input tokens and $3.48 per million output tokens, with V4-Flash even cheaper at $0.14 and $0.28 respectively. A key innovation is its 1-million-token context window, achieved by compressing older information and focusing attention selectively, reducing computing power by 73-90% and memory use by 90-93% compared to its predecessor, V3.2. Notably, V4 is also optimized for domestic Chinese chips like Huawei's Ascend, marking a strategic shift away from Nvidia dependence.

Key takeaway

For AI Architects and NLP Engineers evaluating large language models for deployment, DeepSeek V4 presents a compelling open-source option that balances frontier performance with significantly reduced operational costs, especially for long-context applications. Your teams should consider integrating V4, particularly if you are exploring alternatives to US-made chips or need to process extensive text efficiently, as its pricing is highly competitive and its long-context capabilities are robust.

Key insights

DeepSeek V4 offers competitive performance and cost-efficiency with a 1-million-token context window, optimized for Chinese chips.

Principles

• Selective attention improves long-context efficiency.
• Open-source models can rival closed-source performance.
• Hardware optimization drives cost reduction.

Method

V4's attention mechanism compresses older information while retaining nearby text, significantly reducing computational and memory costs for long context windows, enabling 1-million-token processing.

In practice

• Utilize V4-Pro for complex coding and agent tasks.
• Employ V4-Flash for cost-sensitive, faster applications.
• Explore V4 for applications requiring extensive document analysis.

Topics

• DeepSeek V4
• Open-Source AI Models
• Long Context Windows
• Memory Efficiency
• Chinese AI Chips

Best for: AI Architect, NLP Engineer, Entrepreneur, AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• DeepSeek rolls out new flagship AI model a year after breakthrough - Business Standard — DeepSeek's new V4 models advance open-source AI with enhanced reasoning, agentic tasks, and a million-token context length.
• My Honest Thoughts about Deepseek — DeepSeek V4 offers frontier-level AI capabilities at a fraction of the cost, challenging US AI dominance.
• ChinAI #356: DeepSeek as Road Builder [修路人] — DeepSeek V4 achieves near-frontier AI capabilities with significant compute efficiency, particularly for long-context inference.
• Deepseek v4: Best Opensource Model Ever? (Fully Tested) — Deepseek v4 models offer cost-efficiency and long context, but real-world performance may not match benchmark claims.
• DeepSeek v4: the most expected open-source model ever released, and the quietest landing — DeepSeek v4 prioritizes serving cost reduction through architectural innovation over raw benchmark performance.
• DeepSeek previews new AI model that ‘closes the gap’ with frontier models — DeepSeek V4 models offer competitive performance and unprecedented affordability for large context window MoE architectures.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by MIT Technology Review.