DeepSeek releases V4-Flash model with enhanced capabilities for autonomous agents

DeepSeek unveiled its latest model in the V4 series, featuring substantial improvements to autonomous agent functionality. The 304-billion-parameter model is available at competitive pricing of $0.14 per million input tokens and $0.27 per million output tokens, positioning it favourably in the current market. According to Artificial Analysis benchmarks, the V4-Flash outperforms larger models, including MiniMax's 428-billion-parameter M3. This efficiency in the cost-to-capability ratio positions the model as a potentially superior value option among models available today. Practical testing shows that output quality improves notably when reasoning levels are increased, demonstrating flexibility in performance calibration based on specific use cases.