DeepSeek AI Debuts Open-Source V4 Models With 1M-Token Context

According to a recent LinkedIn post from DeepSeek AI, the company is previewing its new DeepSeek-V4 models with a focus on large 1M-token context windows and cost efficiency. The post highlights two variants, DeepSeek-V4-Pro and DeepSeek-V4-Flash, both positioned as high-performing options relative to leading closed-source models.

The post suggests that DeepSeek-V4-Pro uses a mixture-of-experts architecture with 1.6T total and 49B active parameters, targeting performance at the top end of the market. DeepSeek-V4-Flash, at 284B total and 13B active parameters, is framed as a more economical and faster alternative, potentially broadening adoption among cost-sensitive enterprise and developer users.

As shared in the LinkedIn content, access is available via the company’s chat interface and updated API, and the models’ technical report and open weights are linked publicly. For investors, this open-source strategy could accelerate ecosystem growth, drive developer lock-in, and enhance data flywheels, though it may also limit direct monetization per model and increase reliance on value-added services and infrastructure revenue.

The emphasis on a 1M context length suggests a push into use cases like long-document analysis, codebases, and multi-step workflows that may appeal to enterprise customers seeking to reduce fragmentation across tools. If performance claims hold up in independent benchmarks, the models could strengthen DeepSeek AI’s competitive position against both proprietary and open-source peers, potentially improving its attractiveness as a partner or acquisition target.