Ziqi Ren

Ailin Huang, Ang Li, Aobo Kong et al.

We introduce Step 3.5 Flash, a sparse Mixture-of-Experts (MoE) model that bridges frontier-level agentic intelligence and computational efficiency. We focus on what matters most when building agents: sharp reasoning and fast, reliable execution. Step 3.5 Flash pairs a 196B-parameter foundation with 11B active parameters for efficient inference. It is optimized with interleaved 3:1 sliding-window/full attention and Multi-Token Prediction (MTP-3) to reduce the latency and cost of multi-round agentic interactions. To reach frontier-level intelligence, we design a scalable reinforcement learning framework that combines verifiable signals with preference feedback, while remaining stable under large-scale off-policy training, enabling consistent self-improvement across mathematics, code, and tool use. Step 3.5 Flash demonstrates strong performance across agent, coding, and math tasks, achieving 85.4% on IMO-AnswerBench, 86.4% on LiveCodeBench-v6 (2024.08-2025.05), 88.2% on tau2-Bench, 69.0% on BrowseComp (with context management), and 51.0% on Terminal-Bench 2.0, comparable to frontier models such as GPT-5.2 xHigh and Gemini 3.0 Pro. By redefining the efficiency frontier, Step 3.5 Flash provides a high-density foundation for deploying sophisticated agents in real-world industrial environments.

Chen Hu, Haikuo Du, Heng Wang et al.

As LLMs shift toward autonomous agents, Deep Research has emerged as a pivotal metric. However, existing academic benchmarks like BrowseComp often fail to meet real-world demands for open-ended research, which requires robust skills in intent recognition, long-horizon decision-making, and cross-source verification. To address this, we introduce Step-DeepResearch, a cost-effective, end-to-end agent. We propose a Data Synthesis Strategy Based on Atomic Capabilities to reinforce planning and report writing, combined with a progressive training path from agentic mid-training to SFT and RL. Enhanced by a Checklist-style Judger, this approach significantly improves robustness. Furthermore, to bridge the evaluation gap in the Chinese domain, we establish ADR-Bench for realistic deep research scenarios. Experimental results show that Step-DeepResearch (32B) scores 61.4% on Scale AI Research Rubrics. On ADR-Bench, it significantly outperforms comparable models and rivals SOTA closed-source models like OpenAI and Gemini DeepResearch. These findings prove that refined training enables medium-sized models to achieve expert-level capabilities at industry-leading cost-efficiency.

Lukun Wu, Jie Li, Ziqi Ren et al.

Decoding visual features from EEG signals is a central challenge in neuroscience, with cross-modal alignment as the dominant approach. We argue that the relationship between visual and brain modalities is fundamentally asymmetric, characterized by two critical gaps: a Fidelity Gap (stemming from EEG's inherent noise and signal degradation, vs. vision's high-fidelity features) and a Semantic Gap (arising from EEG's shallow conceptual representation, vs. vision's rich semantic depth). Previous methods often overlook this asymmetry, forcing alignment between the two modalities as if they were equal partners and thereby leading to poor generalization. To address this, we propose the adaptive teaching paradigm. This paradigm empowers the ``teacher" modality (vision) to dynamically shrink and adjust its knowledge structure under task guidance, tailoring its semantically dense features to match the ``student" modality (EEG)'s capacity. We implement this paradigm with the ShrinkAdapter, a simple yet effective module featuring a residual-free design and a bottleneck structure. Through extensive experiments, we validate the underlying rationale and effectiveness of our paradigm. Our method achieves a top-1 accuracy of 60.2\% on the zero-shot brain-to-image retrieval task, surpassing previous state-of-the-art methods by a margin of 9.8\%. Our work introduces a new perspective for asymmetric alignment: the teacher must shrink and adapt to bridge the vision-brain gap.

Ziqi Ren, Jie Li, Xuetong Xue et al.

Reconstructing visual stimulus (image) only from human brain activity measured with functional Magnetic Resonance Imaging (fMRI) is a significant and meaningful task in Human-AI collaboration. However, the inconsistent distribution and representation between fMRI signals and visual images cause the heterogeneity gap. Moreover, the fMRI data is often extremely high-dimensional and contains a lot of visually-irrelevant information. Existing methods generally suffer from these issues so that a satisfactory reconstruction is still challenging. In this paper, we show that it is possible to overcome these challenges by learning visually-guided cognitive latent representations from the fMRI signals, and inversely decoding them to the image stimuli. The resulting framework is called Dual-Variational Autoencoder/ Generative Adversarial Network (D-VAE/GAN), which combines the advantages of adversarial representation learning with knowledge distillation. In addition, we introduce a novel three-stage learning approach which enables the (cognitive) encoder to gradually distill useful knowledge from the paired (visual) encoder during the learning process. Extensive experimental results on both artificial and natural images have demonstrated that our method could achieve surprisingly good results and outperform all other alternatives.