Tsinghua AI

MLLMs possess an intrinsic "S-Space" that acts as a spatial workspace, encoding object coordinates and enabling viewpoint transformations, even under non-spatial queries.

Token-level LLM routing, previously hindered by system inefficiencies, now achieves up to 64x higher throughput with a novel request-centric, model-centric serving architecture.

Irrelevant text distractors cripple mixed-modality retrieval performance more than images, revealing a text modality preference that degrades performance in a V-shaped curve.

Direct regression policies can match generative model performance in robot learning by learning input-dependent residual scales and robustly handling heavy-tailed action distributions.

AI agents can learn to climb leaderboards in complex adversarial games by refining executable policies, but struggle with intricate rule sets

Ambiguous object selection in text-guided image editing is solved by directly annotating the canvas, achieving state-of-the-art performance in complex edits.

Positive Trudinger-Moser solutions on simply connected domains are proven to have bounded Dirichlet energy, precluding critical points at high energy levels.

The Gohberg-Krupnik sharp Volterra conjecture is proven by establishing identical canonical completely bounded norms for the scalar Hilbert transform and the Schur multiplier on Schatten classes.

Decoupling denoising and context writing parameters in autoregressive video models unlocks unprecedented long-horizon generation capabilities, extending continuous output to 24 hours from just 5-second training rollouts.

Network traffic alone can reveal LLM agent risks and granular behaviors, but current analysis methods fail to distinguish subtle malicious activities or semantically similar scenarios.

A unified proactive generation framework enables LLMs to form reusable factual memories from streaming video without compromising real-time perception.

Imagined failures in a world model can pinpoint exactly which robot skill components need new demonstrations, boosting success from 13% to 75% with minimal data.

A rigorous mathematical framework for compressible free-surface fluid dynamics is established, proving local well-posedness for the first time without surface tension.

Legal LLMs can now be optimized for nuanced quality beyond correctness, with rubrics that dissect response evaluation into Style, Element, and Chain of reasoning.

A 260M-parameter looped diffusion model achieves performance surpassing a 6.5x larger non-looped counterpart with significantly less inference compute.

State-of-the-art VLMs struggle to execute scientific research workflows, revealing a critical gap in their ability to interact with specialized software and produce verifiable results.

Robots no longer need human intervention to learn from their mistakes: foundation-model-guided exploration over modular skills can autonomously synthesize corrective rollouts that distill directly into low-level policy weights.

Outcome-only RL can train highly effective process supervision models by rewarding accurate outcome prediction, even without explicit step-level feedback.

Computer-use agents can bypass brittle, hand-engineered APIs entirely by learning when to substitute mouse clicks with native shell commands, yielding a 14.8-point leap on OSWorld.

Video agents can now autonomously discover and fix their own reasoning flaws by re-watching videos and evolving their evidence-gathering strategies.

Today's frontier agents virtually collapse in professional engineering environments, solving just 3.6% of multi-software workflows when held to real-world physical and geometric constraints.

Chain-of-thought token bloat can be slashed by 27% while actually boosting accuracy simply by recognizing that reasoning capacity lives almost entirely within the null space of a base model's dominant singular vectors.

Linear attention's recurrent state can be compressed down to 4 bits while outperforming uniform INT8 once precision is dynamically allocated to long-lived memory tracks and output-sensitive key rows.

Standard pass rates hide severe technical debt: coding agents duplicate logic in over half of multi-turn task chains by turn 5 while progressively abandoning repository exploration.

Runtime scaffolding beats model scale in scientific discovery: structured hypothesis tracking and trajectory branching push DeepSeek-v4-flash from 20.7% to 73.0% accuracy on blind symbolic regression, matching GPT-5.5 without relying on semantic domain clues.

World action models do not need to render clean future video to generalize: conditioning on a single forward pass of pure noise captures virtually all the robustness benefits of multi-step diffusion at a fraction of the compute.

Generative world action models can run up to 3× faster in closed-loop control without fine-tuning or task degradation by dynamically repairing and reusing intermediate computational states instead of recomputing them from scratch.

Discontinuous cost spikes and non-stationary demand jumps do not break analytical tractability: the optimal control for finite-horizon inventory queues collapses to an exact, provably monotonic threshold policy.

Mobile manipulation policies do not need joint action spaces to coordinate effectively: decoupling arm and base streams via masked joint attention and training-time world imagination drives state-of-the-art coordination across 17 distinct embodiments.

Frozen autonomous policies can survive out-of-distribution dynamics without retraining by turning streaming conformal prediction residuals directly into adaptive safety projection margins.

Today's video world models excel at hallucinating plausible-looking frames, but their ability to actually preserve initial object identities drops by over 40% as scene complexity scales beyond 40 instances.

AI scientific agents frequently solve empirical equations for entirely the wrong reasons, failing to identify the correct generative mechanism in over 64% of the cases where they successfully recover the observable phenomenal law.

Distilling causal LLMs into parallel diffusion models fundamentally handicaps students when the teacher ignores visible future tokens—aligning this context yields up to 4-point accuracy gains at 1.5× faster training.

LLMs can scale to full 100,000-word novels without narrative collapse simply by offloading character arcs, plot history, and future constraints to an explicit, training-free state-tracking agent.

A single isolated edge is all that separates linear edge complexity from constant-size collapse in rainbow graph saturation, resolving the asymptotic behavior across all target topologies.

Single-step diffusion models can outperform their multi-step teachers when distributional matching is reformulated as a Wasserstein gradient flow over Gaussian mixtures.

Explicit cross-molecular surface orientations—rather than standard pairwise scalar distances—unlock a 32.5% surge in antibody affinity optimization while cutting CDR structural errors by 15%.

Ground-state preparation query complexity is now strictly settled, proving that the overhead of poor initial overlap $\gamma^{-1}$ and target precision $\log(1/\varepsilon)$ can be decoupled additively rather than multiplicatively in expectation.

Uniform layer looping wastes latent compute on trivial tokens, but dynamically routing iteration depth beats standard transformers on AIME with a 53% steeper test-time scaling slope.

Decades of hand-crafted heuristics for counting and sampling constrained binary matrices can be replaced by a single amortized GFlowNet that achieves a 99.8% median effective sample fraction across unseen margins.

Over 70% of sparse autoencoder features exhibit fundamentally mismatched input and output semantics, breaking the common assumption that what activates a feature directly mirrors its downstream causal effect.

Large reasoning models don't need longer chains of thought to perform better—projecting drifting hidden states back into an efficient latent cluster cuts token generation by up to 53% while boosting task accuracy by up to 12.1%.

Purely autoregressive VLAs no longer require awkward hybrid diffusion heads to achieve fine-grained robot control: structuring discrete action tokenization as causally annealed flow matching recovers continuous diffusion precision while preserving end-to-end token efficiency.

Drawing predicted human gaze crosshairs directly onto input images more than doubles real-world bimanual VLA success (from 26.3% to 56.0%) without requiring runtime eye-tracking hardware or policy architecture changes.

Forcing moving wrist cameras to predict future pixels actively degrades policy performance—decoupling static scene reconstruction from dynamic wrist latent prediction yields up to a 12.8% boost in contact-rich manipulation with zero inference cost.

Diffusion teachers and online score networks are no longer bottlenecks for distilling autoregressive video models: directly minimizing sample MMD against reference videos improves VBench scores while unlocking 14B post-training on just eight H200 GPUs.

Tool-using agents falsely claim success in nearly a quarter of tool failure cases, but enforcing structured evidence contracts slashes these deceptive reporting errors to under 1% while maximizing task recovery.

RoboDawn is investigated, a human-intuitive interface that exposes robotic control to an agentic VLM through a compact set of discrete translation, rotation, and gripper commands, and an in-context learning (ICL) scheme is introduced that ground the VLM in both interface usage and task-solving strategies.

The DeepSeek-V4.1-Flash model, a multimodal Mixture-of-Experts model with 552B backbone parameters and support for contexts of up to one million tokens, is introduced, substantially improving cost efficiency for agentic workloads and pushing the limits of KV cache compression.

The concept of traversability is extended from robot-centric to human-centric by accounting for the operator's cognitive states, such as attention, workload, and risk tolerance or awareness, termed tele-traversability, and is revisited and extended to teleoperation settings.

A synthetic-to-real framework coupling physically validated data generation with a hierarchical, diffusion-based controller is proposed, which outperforms state-of-the-art manipulation policies in transport smoothness and dynamic stability.

The model family shows gains in held-out scientific-code repair and across selected general-purpose benchmarks in code, reasoning, and knowledge, providing evidence of positive transfer from scientific experience to broader capabilities.

Vidu S2, which comprises Vidu S2-Avatar, a real-time interactive digital-character model, and Vidu S2-Editing, a real-time video editing model, is presented and the feasibility of real-time spatial video generation for both Vidu S2-Avatar and Vidu S2-Editing is explored.

Agent memory breaks when models are forced to simultaneously interpret, update, and rewrite facts—decoupling maintenance into explicit pairwise relation classification and role-based fusion yields up to a 29.8 percentage point accuracy leap.

Turning published scientific figures back into executable Python remains a brittle frontier for top VLMs, with even frontier models routinely failing on multi-component layouts, precise axis semantics, and domain typography.

Neural image compression can finally rival classic formats in raw throughput, achieving 2000 FPS decoding and practical 20 FPS encoding while matching JPEG's rate-distortion performance.

Agent memory can be compressed by 50% with virtually zero performance degradation (retaining up to 99.7% accuracy) and a 2x retrieval speedup by structuring historical context into event-centric maximum spanning trees.

GAPrompt++ is proposed, a multi-granular geometry-aware prompting method that provides richer geometric guidance for efficient 3D task adaptation and achieves state-of-the-art performance among prompting-based PEFT methods and even surpasses full fine-tuning across diverse benchmarks, while requiring less than 2% trainable parameters.

Multi-turn jailbreak attacks reveal that framing requests as actionable can significantly reduce model vulnerability, highlighting a critical leverage point in AI safety.

SR-Edit achieves unprecedented image editing fidelity by refining region separation and preserving non-edit areas without introducing new artifacts.

Unpaired data can be leveraged to achieve state-of-the-art performance in solving complex inverse problems, challenging the reliance on paired datasets.

ExBind reveals that even high-validity models can struggle with exact accuracy, exposing critical gaps in visual-to-executable correspondence in AI systems.

A no-code platform that enables the creation of complex multi-agent systems without sacrificing expressiveness or requiring extensive coding.

Evolving causal models allows scientific agents to refine their beliefs and predictions more effectively, outperforming traditional methods in experimental settings.

Time-varying transition dynamics in Poisson-Gamma models lead to a marked improvement in predictive accuracy for count time series.

Current neuron reconstruction methods fall short at the whole-brain scale, revealing critical gaps in structural accuracy that CORAL aims to address.

MatRank's innovative approach to pseudo-labeling reduces prediction error by effectively utilizing unlabeled data, achieving a remarkable NMAE of 0.896.

A closed-loop world model that learns from both successes and failures could revolutionize dexterous manipulation in robotics.

A novel congestion-aware scheduling method reduces delivery times by over 16% in semiconductor fabs without sacrificing throughput.

Recalibrating supervision in world model training can drastically enhance interaction fidelity and physical plausibility without the need for external representations.

Robots can now autonomously evolve their manipulation skills in real-time, leveraging past interactions without needing to update their underlying policy models.

CoMPASS achieves a remarkable balance, enhancing molecular predictions by intelligently integrating LLM insights only when uncertainty is high, leading to improved accuracy without sacrificing reliability.