DeepSeek V4 Flash API adds 1M context, thinking modes, and tool calls

DeepSeek has quietly updated its API documentation with details on the V4 Flash model, which introduces significant flexibility for developers. The model supports both thinking and non-thinking inference modes, JSON-structured output, and tool-calling capabilities—all within a massive 1 million token context window. This positions V4 Flash as a versatile option for tasks requiring deep reasoning, structured data extraction, or agentic workflows, similar to frontier models but at presumably lower cost.

Meanwhile, the AI community on Reddit is buzzing about hints of an imminent DeepSeek V4.1 update, possibly dropping as early as this week. Some API users have reported intermittent speed bumps, which could indicate backend changes or preparation for the new version. If V4.1 follows the trend of rapid iteration seen with earlier DeepSeek releases, it may bring further improvements in inference speed or accuracy. For professionals, this signals that DeepSeek remains highly active in the competitive AI model space, offering increasingly capable models for production workloads.