DeepSeek previews new AI model that ‘closes the gap’ with frontier models

2026-04-24 · Source: AI News & Artificial Intelligence | TechCrunch · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Advanced, quick

Summary

DeepSeek, a Chinese AI lab, has released preview versions of its new large language models, DeepSeek V4 Flash and V4 Pro, which are updates to their V3.2 model. Both are mixture-of-experts models featuring a 1 million token context window, designed to handle extensive codebases or documents. The V4 Pro model is the largest open-weight model available, with 1.6 trillion parameters (49 billion active), surpassing Moonshot AI’s Kimi K 2.6 and MiniMax’s M1. The smaller V4 Flash has 284 billion parameters (13 billion active). DeepSeek claims these models are more efficient and performant than V3.2, with V4-Pro-Max outperforming open-source peers and some OpenAI and Gemini models on reasoning and coding benchmarks, though they lag in knowledge tests. The models are text-only and significantly more affordable than current frontier models, with V4 Flash costing $0.14 per million input tokens and V4 Pro costing $0.145 per million input tokens.

Key takeaway

For AI Engineers evaluating large language models for cost-sensitive applications, DeepSeek V4 Flash and V4 Pro present a compelling option. Their 1 million token context window and significantly lower pricing per million tokens, compared to leading frontier models, make them ideal for tasks requiring extensive input processing where multimodal capabilities are not critical. Consider benchmarking these models for reasoning and coding tasks to validate performance for your specific use cases.

Key insights

DeepSeek V4 models offer competitive performance and unprecedented affordability for large context window MoE architectures.

Principles

• Mixture-of-experts lowers inference costs.
• Larger context windows enable extensive input processing.

In practice

• Utilize 1M token context for large codebases.
• Employ MoE models to manage inference expenses.

Topics

• DeepSeek V4
• Mixture-of-Experts
• Large Language Models
• AI Benchmarks
• Open-weight Models

Best for: AI Engineer, NLP Engineer, CTO, AI Scientist, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• DeepSeek V4 - almost on the frontier, a fraction of the price — DeepSeek V4 models offer competitive performance and industry-leading low costs through significant efficiency gains.
• 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.
• DeepSeek V4: One Million Tokens, Three Thinking Modes, and the First Real Hands-On Reports — DeepSeek-V4 optimizes long-context LLMs via novel attention and training stability, offering tiered reasoning and cost-effective variants.
• DeepSeek v4 Pro/Flash - Benchmarks and OpenCode Test | Frontend, SVG, GameDev, Backend | 🔴 Live — Deepseek V4 introduces large Mixture-of-Experts models with an efficient attention mechanism, showing mixed performance and API reliability.
• As agentic AI pushes rivals to raise prices and cap usage, Deepseek ships a good-enough model for almost nothing — Deepseek's V4 models offer massive context windows and competitive performance at significantly reduced costs via architectural innovation.
• Three reasons why DeepSeek’s new model matters — DeepSeek V4 offers competitive performance and cost-efficiency with a 1-million-token context window, optimized for Chinese chips.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI News & Artificial Intelligence | TechCrunch.