Study: DeepSeek V4-Flash runs at least 1% of Claude Fable 5

On 3 August, data from the research institute Artificial Analysis showed that DeepSeek ' s flagship artificial intelligence model, V4-Flash, had the lowest operating cost in the baseline test of the main model and less than 1 per cent of Claude Fable 5 of the Anthropic flagship model. Artificial Analysis states that V4-Flash charges $0.14 per 1 million input characters (Token) and $0.28 per 1 million output terms. According to the Agency ' s estimates, the average cost of V4-Flash to complete each test was only 3 cents, well below 86 cents of Kimi K3, OpenAI GPT-5.6 Sol and $3.15 of Claude Fable 5. In its "Intelligence Index" rating, V4-Flash scored 50, divided into 100 points, the same as Gemini 3.6 Flash under the Google flag, one point lower than Muse Spark 1.1 and GLM-5.2 in Meta. However, Kimi K3 received 57 points on the dark side of the month; Claude Opus 5 and Claude Fable 5 in Anthropic and OpenAI GPT-5.6 scored at least 9 points higher than V4-Flash. DeepSeek officially released V4-Flash last Friday. DeepSeek states that the V4-Flash is now in the OpenAI Responses API format and fully compatible with OpenAI ' s artificially intelligent code proxy tool Codex, and that developers can use the lower cost V4-Flash to perform such tasks as code generation and proxy development。