DeepSeek V4-Flash Crushes Competitors in Cost-Performance Ratio, Operating Costs Only 1/105 of Claude

According to TechFlow from DeepSeek, on August 3rd, Reuters reported that Chinese AI startup DeepSeek officially launched the latest version of its API model, V4-Flash, on July 31st. In benchmark tests conducted by AI performance analysis agency Artificial Analysis, the operating cost of V4-Flash was only 1/105 of Anthropic Claude Fable 5. In terms of specific pricing, the input token fee for V4-Flash is $0.14 per million tokens, while the output token fee is $0.28 per million tokens, with an average test cost of approximately $0.03—significantly lower than Wenxin Kimi K3 ($0.86), OpenAI GPT-5.6 Sol ($1.86), and Claude Fable 5 ($3.15). In performance metrics, V4-Flash scored 50 points on the comprehensive intelligence index, tying with Google Gemini 3.6 Flash but still trailing leading models like Claude Opus 5 and GPT-5.6 by over 9 points. It is worth noting that low pricing does not necessarily equate to low actual costs—if the model requires more inference and output tokens to generate responses, the actual expenses could increase significantly.