DeepSeek-V4-Flash-0731 Released With Enhanced Agentic Capabilities

DeepSeek-V4-Flash-0731 is the newest artificial intelligence model in the DeepSeek V4 family, featuring substantially enhanced agentic capabilities. The model contains 304 billion parameters and takes up 167 gigabytes on Hugging Face.

Despite its size, the model performs strongly compared to larger alternatives. Artificial Analysis ranks it ahead of the 428-billion-parameter MiniMax M3 model. The pricing is set at fourteen cents per million input tokens and twenty-seven cents per million output tokens, positioning it as an exceptionally high-value option relative to its intelligence level.

Performance on specific tasks can vary based on configuration settings. For instance, generating certain image descriptions with default reasoning levels yields poor results, but increasing the reasoning effort to high significantly improves the output.

This release matters because it delivers high-level performance and advanced agentic features at a very low cost, making it one of the most cost-effective artificial intelligence models currently available on the market.