Three reasons why DeepSeek’s new model matters

Chinese AI firm DeepSeek released V4, its long-awaited flagship open-source model, on Friday. V4 comes in two versions: V4-Pro, a larger model built for coding and complex agent tasks, and V4-Flash, a smaller, faster, and cheaper version for general use. Both handle 1 million tokens—enough for all three volumes of The Lord of the Rings and The Hobbit combined. DeepSeek claims V4-Pro rivals top closed-source models like Anthropic's Claude-Opus-4.6, OpenAI's GPT-5.4, and Google's Gemini-3.1 on major benchmarks, and exceeds other open-source models like Alibaba's Qwen-3.5 on coding, math, and STEM tasks. Pricing is aggressive: V4-Pro costs $1.74 per million input tokens and $3.48 per million output tokens, while V4-Flash costs just $0.14 and $0.28 respectively—a fraction of OpenAI and Anthropic's rates. In an internal survey of 85 developers, over 90% ranked V4-Pro among their top choices for coding tasks. DeepSeek optimized V4 for agent frameworks like Claude Code, OpenClaw, and CodeBuddy, making it a strong option for building applications.

This release is DeepSeek's most significant since R1 in January 2025, which stunned the industry with its efficiency on limited compute. V4 introduces a new design for memory efficiency, enabling its long context window—now the default across all DeepSeek services, matching cutting-edge models from Google and Anthropic. While V4 is unlikely to shake the AI field as dramatically as R1 did, it solidifies DeepSeek's position as a top-tier open-source provider. The model is available on DeepSeek's website and app, with API access open to developers. This comes after months of scrutiny for the company, including personnel departures and delays, but V4 demonstrates DeepSeek's continued ability to deliver frontier AI capabilities at a fraction of the cost.