Step 3.7 Flash Achieves 409 Tokens/s, Tops Artificial Analysis Output Speed Ranking

ME AI message: According to the latest Output Speed ranking from Artificial Analysis, the globally respected large model evaluation platform, StepFun’s newly open-sourced base model, Step 3.7 Flash, leads all mainstream models with an output speed of 409 tokens/s. It also holds a leading position across key metrics including End-to-End Response Time, Intelligence vs. Output Speed, and Output Speed vs. Price. From an industry perspective, competition among large models is shifting from isolated capability benchmarks to real-world Agent task efficiency. In complete task pipelines involving browsing, retrieval, document understanding, interface analysis, and tool invocation, models are no longer merely answering questions—they function as continuous execution engines. As a result, end-to-end latency, throughput capacity, and cost structure have become critical constraints. Therefore, a systemic balance of higher throughput, lower latency, and superior cost efficiency is emerging as the foundational requirement for scalable Agent deployment. In this context, Step 3.7 Flash outperforms comparable models across multiple dimensions—including Intelligence vs. Output Speed, End-to-End Response Time, and Output Speed vs. Price—achieving coordinated optimization of intelligence, speed, and cost. This provides essential foundational support for Agent systems requiring high-frequency invocation, continuous operation, and scalable deployment. This trend further confirms that the core competitiveness in the Agent era is shifting from peak model capability to real-world task completion efficiency—essentially, a systemic balance among speed, intelligence, and cost. Meanwhile, since its release, Step 3.7 Flash has risen to second place on OpenRouter’s Trending list and has become one of the most closely watched open-source models in the global developer community. Developer feedback highlights its outstanding performance in runtime efficiency, multimodal understanding, and Agent tool invocation capabilities. Some developers have compared it with similar models like DeepSeek V4 Flash and concluded that it offers clear advantages in speed and response experience. Overall, Step 3.7 Flash’s dual recognition in authoritative evaluations and the developer community not only validates its engineering strengths in high throughput and low latency but also reflects how Chinese open-source models are rapidly integrating into the global developer ecosystem and securing a more prominent role in the next phase of Agent infrastructure competition. (Source: Ifnar)