According to Artificial Analysis's evaluation, DeepSeek V4 Flash achieved an intelligence index of 50, ranking third among 101 similar models, more than double the median of open-weight models. Its output speed is approximately 104 tokens per second,

2026-08-06

According to Artificial Analysis's evaluation, DeepSeek V4 Flash achieved an intelligence index of 50, ranking third among 101 similar models, more than double the median of open-weight models. Its output speed is approximately 104 tokens per second, with input and output prices of only $0.14 and $0.28 per million tokens, respectively. This indicates that model competition is shifting from simply pursuing the highest performance to comparing "how much it costs to complete a task." The low price, open weights, and million-token context make it more suitable for batch searches, programming assistance, and multi-agent invocation, forcing high-priced closed-source models to maintain their premium through reliability, ecosystem, and enterprise services. However, DeepSeek generated 210 million tokens in the evaluation, approximately twice the median of similar models, indicating its relatively lengthy inference process. Benchmark scores also cannot fully reflect the success rate of complex tasks; therefore, the low token price does not mean that all real-world tasks will achieve the same proportional cost advantage.