Developing embodied AI agents requires scalable training environments that balance content diversity with physics accuracy. World simulators provide such environments but face distinct limitations: video-based methods generate diverse content but lack real-time physics feedback for interactive learning, while physics-based engines provide accurate dynamics but face scalability limitations from costly manual asset creation. We present Seed3D 1.0, a foundation model that generates simulation-ready 3D assets from single images, addressing the scalability challenge while maintaining physics rigor. Unlike existing 3D generation models, our system produces assets with accurate geometry, well-aligned textures, and realistic physically-based materials. These assets can be directly integrated into physics engines with minimal configuration, enabling deployment in robotic manipulation and simulation training. Beyond individual objects, the system scales to complete scene generation through assembling objects into coherent environments. By enabling scalable simulation-ready content creation, Seed3D 1.0 provides a foundation for advancing physics-based world simulators.

Predictive simulation of surface chemistry is critical in fields from catalysis to electrochemistry and clean energy generation. Ab-initio quantum many-body methods should offer deep insights into these systems at the electronic level but are limited by their steep computational cost. Here, we build upon state-of-the-art correlated wavefunctions to reliably reach ‘gold standard’ accuracy in quantum chemistry for extended surface chemistry. Efficiently harnessing graphics processing unit acceleration along with systematically improvable multi-resolution techniques, we achieve linear computational scaling up to 392 atoms. These large-scale simulations demonstrate the importance of converging to these extended system sizes, achieving consistency between simulations with different boundary conditions for the interaction of water on a graphene surface. We provide a benchmark for this water-graphene interaction that clarifies the preference for water orientations at the graphene interface. This is extended to the adsorption of carbonaceous molecules on chemically complex surfaces, including metal oxides and metal-organic frameworks, where we consistently achieve chemical accuracy compared to experimental references. This advances the simulation of molecular adsorption on surfaces, enabling reliable and improvable first-principles modeling of such problems by ab-initio quantum many-body methods.

The remarkable success of large language models (LLMs) stems from their ability to consolidate vast amounts of knowledge into the memory during pre-training and to retrieve it from the memory during inference, enabling advanced capabilities such as knowledge memorization, instruction-following and reasoning. However, the mechanisms of memory retrieval and consolidation in LLMs remain poorly understood. In this paper, we propose the function token hypothesis to explain the workings of LLMs: During inference, function tokens activate the most predictive features from context and govern next token prediction (memory retrieval). During pre-training, predicting the next tokens (usually content tokens) that follow function tokens increases the number of learned features of LLMs and updates the model parameters (memory consolidation). Function tokens here roughly correspond to function words in linguistics, including punctuation marks, articles, prepositions, and conjunctions, in contrast to content tokens. We provide extensive experimental evidence supporting this hypothesis. Using bipartite graph analysis, we show that a small number of function tokens activate the majority of features. Case studies further reveal how function tokens activate the most predictive features from context to direct next token prediction. We also find that during pre-training, the training loss is dominated by predicting the next content tokens following function tokens, which forces the function tokens to select the most predictive features from context.

Rapid advances in text-to-image (T2I) generation have raised higher requirements for evaluation methodologies. Existing benchmarks center on objective capabilities and dimensions, but lack an application-scenario perspective, limiting external validity. Moreover, current evaluations typically rely on either ELO for overall ranking or MOS for dimension-specific scoring, yet both methods have inherent shortcomings and limited interpretability. Therefore, we introduce the Magic Evaluation Framework (MEF), a systematic and practical approach for evaluating T2I models. First, we propose a structured taxonomy encompassing user scenarios, elements, element compositions, and text expression forms to construct the Magic-Bench-377, which supports label-level assessment and ensures a balanced coverage of both user scenarios and capabilities. On this basis, we combine ELO and dimension-specific MOS to generate model rankings and fine-grained assessments respectively. This joint evaluation method further enables us to quantitatively analyze the contribution of each dimension to user satisfaction using multivariate logistic regression. By applying MEF to current T2I models, we obtain a leaderboard and key characteristics of the leading models. We release our evaluation framework and make Magic-Bench-377 fully open-source to advance research in the evaluation of visual generative models.

This research introduces ByteWrist, a novel highly-flexible and anthropomorphic parallel wrist for robotic manipulation. ByteWrist addresses the critical limitations of existing serial and parallel wrists in narrow-space operations through a compact three-stage parallel drive mechanism integrated with arc-shaped end linkages. The design achieves precise RPY (Roll-Pitch-Yaw) motion while maintaining exceptional compactness, making it particularly suitable for complex unstructured environments such as home services, medical assistance, and precision assembly. The key innovations include: (1) a nested three-stage motor-driven linkages that minimize volume while enabling independent multi-DOF control, (2) arc-shaped end linkages that optimize force transmission and expand motion range, and (3) a central supporting ball functioning as a spherical joint that enhances structural stiffness without compromising flexibility. Meanwhile, we present comprehensive kinematic modeling including forward / inverse kinematics and a numerical Jacobian solution for precise control. Empirically, we observe ByteWrist demonstrates strong performance in narrow-space maneuverability and dual-arm cooperative manipulation tasks, outperforming Kinova-based systems. Results indicate significant improvements in compactness, efficiency, and stiffness compared to traditional designs, establishing ByteWrist as a promising solution for next-generation robotic manipulation in constrained environments.

PXDesign achieves nanomolar binder hit rates of 20–73% across five of six diverse protein targets, surpassing prior methods such as AlphaProteo. This experimental success rate is enabled by advances in both binder generation and filtering. We develop both a diffusion-based generative model (PXDesign-d) and a hallucination-based approach (PXDesign-h), each showing strong in silico performance that outperforms existing models. Beyond generation, we systematically analyze confidence-based filtering and ranking strategies from multiple structure predictors, comparing their accuracy, efficiency, and complementarity on datasets spanning de novo binders and mutagenesis. Finally, we validate the full design process experimentally, achieving high hit rates and multiple nanomolar binders. To support future work and community use, we release a unified benchmarking framework at https://github.com/bytedance/PXDesignBench, provide public access to PXDesign via a webserver at https://protenix-server.com, and share all designed binder sequences at https://protenix.github.io/pxdesign.

We introduce Robix, a unified vision-language model designed to serve as the high-level cognitive layer in a hierarchical robot system, integrating robot reasoning, task planning, and natural language interaction within a single architecture. Robix dynamically generates atomic commands for low-level controllers alongside verbal responses for human interaction, enabling end-to-end execution of complex instructions, long-horizon task planning, and natural human-robot collaboration. The model also introduces novel capabilities such as proactive dialogue, real-time interruption handling, and context-aware commonsense reasoning during task execution. At its core, Robix employs chain-of-thought reasoning and is trained through a three-stage strategy: (1) continued pretraining to enhance embodied reasoning skills like 3D spatial understanding, visual grounding, and task-centric reasoning; (2) supervised finetuning to model human-robot interaction and task planning as a unified reasoning-action sequence; and (3) reinforcement learning to improve reasoning-action consistency and long-horizon task coherence. Extensive experiments show that Robix outperforms both open-source and commercial baselines—including GPT-4o and Gemini 2.5 Pro—in interactive task execution, demonstrating strong generalization across diverse instruction types (e.g., open-ended, multi-stage, constrained, invalid, and interrupted) and various user-involved tasks such as table bussing, grocery shopping, and dietary filtering.

We introduce M3-Agent, a novel multimodal agent framework equipped with long-term memory. Like humans, M3-Agent can process real-time visual and auditory inputs to build and update its long-term memory. Beyond episodic memory, it also develops semantic memory, enabling it to accumulate world knowledge over time. Its memory is organized in an entity-centric, multimodal format, allowing deeper and more consistent understanding of the environment. Given an instruction, M3-Agent autonomously performs multi-turn, iterative reasoning and retrieves relevant information from memory to accomplish the task. To evaluate memory effectiveness and memory-based reasoning in multimodal agents, we develop M3-Bench, a new long-video question answering benchmark. M3-Bench comprises 100 newly recorded real-world videos captured from a robot's perspective (M3-Bench-robot) and 929 web-sourced videos across diverse scenarios (M3-Bench-web). We annotate question-answer pairs designed to test key capabilities essential for agent applications, such as human understanding, general knowledge extraction, and cross-modal reasoning. Experimental results show that M3-Agent, trained via reinforcement learning, outperforms the strongest baseline, a prompting agent using Gemini-1.5-pro and GPT-4o, achieving 6.7%, 7.7%, and 5.3% higher accuracy on M3-Bench-robot, M3-Bench-web and VideoMME-long, respectively. Our work advances the multimodal agents toward more human-like long-term memory and provides insights into their practical design. Model, code and data are available at https://github.com/bytedance-seed/m3-agent.

We present Seed Diffusion Preview, a large-scale language model based on discrete-state diffusion, offering remarkably fast inference speed. Thanks to non-sequential, parallel generation, discrete diffusion models provide a notable speedup to mitigate the inherent latency of token-by-token decoding, as demonstrated recently (e.g., Mercury Coder, Gemini Diffusion). Seed Diffusion Preview achieves an inference speed of 2,146 token/s over H20 GPUs while maintaining competitive performance across a sweep of standard code evaluation benchmarks, significantly faster than contemporary Mercury and Gemini Diffusion, establishing new state of the art on the speed-quality Pareto frontier for code models.

LLMs have demonstrated strong mathematical reasoning abilities by leveraging reinforcement learning with long chain-of-thought, yet they continue to struggle with theorem proving due to the lack of clear supervision signals when solely using natural language. Dedicated domain-specific languages like Lean provide clear supervision via formal verification of proofs, enabling effective training through reinforcement learning. In this work, we propose \textbf{Seed-Prover}, a lemma-style whole-proof reasoning model. Seed-Prover can iteratively refine its proof based on Lean feedback, proved lemmas, and self-summarization. To solve IMO-level contest problems, we design three test-time inference strategies that enable both deep and broad reasoning. Seed-Prover proves 78.1% of formalized past IMO problems, saturates MiniF2F, and achieves over 50\% on PutnamBench, outperforming the previous state-of-the-art by a large margin. To address the lack of geometry support in Lean, we introduce a geometry reasoning engine \textbf{Seed-Geometry}, which outperforms previous formal geometry engines. We use these two systems to participate in IMO 2025 and fully prove 5 out of 6 problems. This work represents a significant advancement in automated mathematical reasoning, demonstrating the effectiveness of formal verification with long chain-of-thought reasoning.

Simultaneous Interpretation (SI) represents one of the most daunting frontiers in the translation industry, with product-level automatic systems long plagued by intractable challenges: subpar transcription and translation quality, lack of real-time speech generation, multi-speaker confusion, and translated speech inflation, especially in long-form discourses. In this study, we introduce Seed-LiveInterpret 2.0, an end-to-end SI model that delivers high-fidelity, ultra-low-latency speech-to-speech generation with voice cloning capabilities. As a fully operational product-level solution, Seed-LiveInterpret 2.0 tackles these challenges head-on through our novel duplex speech-to-speech understanding-generating framework. Experimental results demonstrate that through large-scale pretraining and reinforcement learning, the model achieves a significantly better balance between translation accuracy and latency, validated by human interpreters to exceed 70% correctness in complex scenarios. Notably, Seed-LiveInterpret 2.0 outperforms commercial SI solutions by significant margins in translation quality, while slashing the average latency of cloned speech from nearly 10 seconds to a near-real-time 3 seconds, which is around a near 70% reduction that drastically enhances practical usability.

We report our recent progress towards building generalist robot policies, the development of GR-3. GR-3 is a large-scale vision-language-action (VLA) model. It showcases exceptional capabilities in generalizing to novel objects, environments, and instructions involving abstract concepts. Furthermore, it can be efficiently fine-tuned with minimal human trajectory data, enabling rapid and cost-effective adaptation to new settings. GR-3 also excels in handling long-horizon and dexterous tasks, including those requiring bi-manual manipulation and mobile movement, showcasing robust and reliable performance. These capabilities are achieved through a multi-faceted training recipe that includes co-training with web-scale vision-language data, efficient fine-tuning from human trajectory data collected via VR devices, and effective imitation learning with robot trajectory data. In addition, we introduce ByteMini, a versatile bi-manual mobile robot designed with exceptional flexibility and reliability, capable of accomplishing a wide range of tasks when integrated with GR-3. Through extensive real-world experiments, we show GR-3 surpasses the state-of-the-art baseline method, π0, on a wide variety of challenging tasks. We hope GR-3 can serve as a step towards building generalist robots capable of assisting humans in daily life.

Large Language Models (LLMs) have shown impressive performance in complex reasoning tasks through the use of Chain-of-Thought (CoT) reasoning, allowing models to break down problems into manageable sub-tasks. However, existing CoT evaluation techniques either require annotated CoT data or fall short in accurately assessing intermediate reasoning steps, leading to high rates of false positives. In this paper, we formalize CoT reasoning in LLMs through an information-theoretic lens. Specifically, our framework quantifies the 'information-gain' at each reasoning step, enabling the identification of failure modes in LLMs without the need for expensive annotated datasets. We demonstrate the efficacy of our approach through extensive experiments on toy arithmetic, GSM8K and PRM800k datasets, where it significantly outperforms existing outcome-based methods by providing more accurate insights into model performance on individual subtasks.

Replicating human--level dexterity remains a fundamental robotics challenge, requiring integrated solutions from mechatronic design to the control of high degree--of--freedom (DoF) robotic hands. While imitation learning shows promise in transferring human dexterity to robots, the efficacy of trained policies relies on the quality of human demonstration data. We bridge this gap with a hand--arm teleoperation system featuring: (1) a 20--DoF linkage--driven anthropomorphic robotic hand for biomimetic dexterity, and (2) an optimization--based motion retargeting for real--time, high--fidelity reproduction of intricate human hand motions and seamless hand--arm coordination. We validate the system via extensive empirical evaluations, including dexterous in-hand manipulation tasks and a long--horizon task requiring the organization of a cluttered makeup table randomly populated with nine objects. Experimental results demonstrate its intuitive teleoperation interface with real--time control and the ability to generate high--quality demonstration data. Please refer to the accompanying video for further details.

It is well-known that the Hessian of deep loss landscape matters to optimization and generalization of deep learning. Previous studies reported a rough Hessian structure in deep learning, which consists of two components, a small number of large eigenvalues and a large number of nearly-zero eigenvalues. To the best of our knowledge, we are the first to report that a simple but overlooked power-law Hessian structure exists in well-trained deep neural networks, including Convolutional Neural Networks (CNNs) and Large Language Models (LLMs). Moreover, we provide a maximum-entropy theoretical interpretation for the power-law Hessian structure and theoretically demonstrate the existence of a robust and low-dimensional subspace of deep neural networks. Our extensive experiments using the proposed power-law spectral method demonstrate that the power-law Hessian spectra critically relate to multiple important behaviors of deep learning, including optimization, generalization, and overparameterization. Notably, we discover that the power-law Hessian structure of a given LLM can often predict generalization during training in some occasions, while conventional sharpnessbased generalization measures which often work well on CNNs largely fail as an effective generalization predictor of LLMs.

Building neural reward models from human preferences is a pivotal component in reinforcement learning from human feedback (RLHF) and large language model alignment research. Given the scarcity and high cost of human annotation, how to select the most informative pairs to annotate is an essential yet challenging open problem. In this work, we highlight the insight that an ideal comparison dataset for reward modeling should balance exploration of the representation space and make informative comparisons between pairs with moderate reward differences. Technically, challenges arise in quantifying the two objectives and efficiently prioritizing the comparisons to be annotated. To address this, we propose the Fisher information-based selection strategies, adapt theories from the classical experimental design literature, and apply them to the final linear layer of the deep neural network-based reward modeling tasks. Empirically, our method demonstrates remarkable performance, high computational efficiency, and stability compared to other selection methods from deep learning and classical statistical literature across multiple open-source LLMs and datasets. Further ablation studies reveal that incorporating cross-prompt comparisons in active reward modeling significantly enhances labeling efficiency, shedding light on the potential for improved annotation strategies in RLHF. Code and embeddings to reproduce all results of this paper are available at https://github.com/YunyiShen/ARM-FI/.

Inference latency stands as a critical bottleneck in the large-scale deployment of Large Language Models (LLMs). Speculative decoding methods have recently shown promise in accelerating inference without compromising the output distribution. However, existing work typically relies on a dualistic draft-verify framework and lacks rigorous theoretical grounding. In this paper, we introduce a novel polybasic speculative decoding framework, underpinned by a comprehensive theoretical analysis. Specifically, we prove a fundamental theorem that characterizes the optimal inference time for multi-model speculative decoding systems, shedding light on how to extend beyond the dualistic approach to a more general polybasic paradigm. Through our theoretical investigation of multi-model token generation, we expose and optimize the interplay between model capabilities, acceptance lengths, and overall computational cost. Our framework supports both standalone implementation and integration with existing speculative techniques, leading to accelerated performance in practice. Experimental results across multiple model families demonstrate that our approach yields speedup ratios ranging from 3.31× to 4.01× for LLaMA2-Chat 7B, up to 3.87× for LLaMA3-8B, up to 4.43× for Vicuna7B and up to 3.85× for Qwen2-7B—all while preserving the original output distribution. We release our theoretical proofs and implementation code to facilitate further investigation into polybasic speculative decoding.

We propose an imperceptible multi-bit text watermark embedded by paraphrasing with LLMs. We fine-tune a pair of LLM paraphrasers that are designed to behave differently so that their paraphrasing difference reflected in the text semantics can be identified by a trained decoder. To embed our multi-bit watermark, we use two paraphrasers alternatively to encode the pre-defined binary code at the sentence level. Then we use a text classifier as the decoder to decode each bit of the watermark. Through extensive experiments, we show that our watermarks can achieve over 99.99% detection AUC with small (1.1B) text paraphrasers while keeping the semantic information of the original sentence. More importantly, our pipeline is robust under word substitution and sentence paraphrasing perturbations and generalizes well to out-of-distributional data. We also show the stealthiness of our watermark with LLM-based evaluation. We open-source the code: https://github.com/xiaojunxu/multi-bit-text-watermark.

Vision-Language Models (VLMs) such as CLIP excel at zero-shot classification due to large-scale pre-training but are vulnerable to adversarial examples. Adversarial fine-tuning robustifies zero-shot models by aligning prediction scores of individual adversaries with their clean counterparts, which typically overlooks intermediate adversarial samples along the adversarial trajectory crossing the decision boundary. Such intermediate adversaries and their vicinity produce informative representations capturing the decision boundary in detail. They can be improved by sampling adversarial candidates from simplices formed by joining two consecutive vertices on the adversarial trajectory and their clean counterpart. However, sampling simplices for adversaries is very costly. To train robust VLM, we overcome these limitations by Taylor expansion and formulating an upper-bound of alignment loss that depends on the Jacobian/Hessian obtained at clean samples. As regions between clean and intermediate adversarial samples capture a larger decision landscape, we robustify VLM by plausible adversaries from simplices by our closed-form formulation equivalent to infinite uniform sampling of the simplex. We obtain state-of-the-art robustness across 15 datasets and diverse vision-language tasks.