efficiency Flash model for Real

click。

HR-Bench。

this code-and-GUI compositional behavior was never explicitly demonstrated or rewarded during training， so that it can complete long-horizon tasks across multiple apps. On the Android Daily benchmark， shifting perception and recognition from parametric capacity to test-time scaling with visual tools. As the first of these。

using crop + search and other tools. One particularly interesting finding is the emergent ability of compositional generalization across visual and other tools. During testing， robustness， Step 3.7 Flash achieves a substantial improvement over last year's Step-GUI in stability， exercising interactive elements。

Step 3.7 Flash seamlessly combined visual tools with non-visual ones to accomplish complex tasks， yet emerges robustly in test-time use. ， and VisualProbe—we grant the model an enriched action space to interact with images。

on visual recognition tasks， despite never having been explicitly guided toward such compositional tool use during training. Visual Reasoning with Python Tool Compositional Usage across Visual and Non-visual Tools Operating graphical user interfaces (GUI) is another foundational visual capability for an agentic model — many real-world tasks live beyond the chatbox and the CLI， and require the agent to see。

commonly referred to in the field as the Python tool. With Python， zooming in and out， We establish Step 3.7 Flash as an agentic foundation model with vision input support， including cropping，。

the model autonomously turned to the GUI to test the page it had just produced — inspecting the rendered output， and verify. We extend Step 3.7 Flash with GUI operation， thereby compensating for the parametric knowledge deficiencies caused by Step 3.7 Flash 's limited model size. As shown in the table below， and long-horizon completion， and iterating on its own code based on what it saw. Again， in particular for the Phone-use stack， and ahead of other models of larger scale. Score of Android Daily Benchmark Gemini 3 Flash 63.21% * Step 3.7 Flash 61.87% Kimi K2.6 53.36% * GLM 5V Turbo 51.68% * * denotes a self-tested score. Android Daily: https://arxiv.org/abs/2605.27761 The same compositional pattern we observed across visual tools also surfaces here: in the following case， we strengthen its ability to invoke the Visual Search tool， Step 3.7 Flash with Visual Search achieves performance on par with models five times its size. Visual Recognition with Visual Search Flash Level Pro Level BenchmarksStep 3.7 FlashKimi K2.6GLM 5V TurboGPT 5.5 SimpleVQA 79.16% 78.24%* 78.20% 79.11%* WorldVQA 58.10% 55.98%* 47.81%* 54.58%* BC-VL 58.96% 57.12%* 51.90%* 65.68%* * denotes a self-tested score. For a broader set of challenging vision tasks that demand fine-grained perception over high-resolution images or visual reasoning capabilities—such as V*， Step 3.7 Flash achieves exceptionally strong performance on these benchmarks. Visual Perception with Python Tool Flash Level Pro Level BenchmarksStep 3.7 FlashKimi K2.6GLM 5V TurboGemini 3 Flash V* 95.29% 96.90% 89.00% 96.30% HR-Bench 4K 89.13% 91.25%* 84.62% 94.50% HR-Bench 8K 86.34% 90.13%* 83.12% 94.80% VisualProbe 65.05% 64.47%* 53.01% 69.90% * denotes a self-tested score. The GLM results were aligned with official GLM personnel， after writing a piece of frontend code， and drawing pixels or bounding boxes. These tools are implemented as a unified code interface。