Reachability Is Not Realization: Tracing the Sources of LLM Benchmark Gains

Benchmark gains are often treated as evidence of greater LLM capability. Yet the same gain can reflect different changes in model behavior. A model may reach new answers, or produce answers that were already within reach. Aggregate scores do not distinguish these changes question by question. We establish a question-level audit under fixed budgets, temperatures, and answer formats. A question is realized when the default deployment procedure produces the correct answer. A question is reachable when a specified probe finds that answer within a fixed budget. We first test whether inference-time layer routing can expand reachability. Under a matched budget, random routes match or exceed structured search in all 43 model and task settings. Answer-blind procedures retain almost none of this gain, which instead requires access to the correct answer. We then ask why reachable answers sometimes fail to appear. Across six cases spanning 0.5B to 31B, silencing one identified MLP block repairs 68 to 92 percent of a predefined failure set. We next test whether training closes the gap by expanding reachability. In five of six matched evaluations, deployed performance rises while the reachable ceiling remains flat or falls. For DAPO, the deployed score rises by 14.7 points while the reachable ceiling falls by 13.3 points. Across the settings we audit, realization and reachability therefore do not always change together. Claims of capability expansion should report both realized performance and reachability under matched evaluation conditions. Code is available at https://github.com/LiZaiyuan0619/reachability-not-realization

Disagree to Accelerate: Closing the Loop on Diffusion Feature Forecasts

Training-free feature forecasting accelerates diffusion sampling by predicting features at skipped denoising steps. Recent work has mainly focused on designing stronger forecasters. Yet forecast error varies sharply across steps, and open-loop caches trust the forecast in full at every skipped step. This fixed trust is what breaks as acceleration turns aggressive. The missing question is not only how to forecast better, but when and how much to trust a forecast. We show that reliability can be observed from the cache itself. Two forecasts agree where the feature trajectory is smooth, and they diverge where prediction turns hard. Their disagreement is a cheap runtime signal, and it costs no extra denoiser evaluation. Based on this signal, we introduce RACER, a training-free closed-loop controller with two responses. It continuously shrinks uncertain forecasts toward the last computed feature. At the riskiest steps, RACER refreshes the feature and repays the added evaluation by skipping a later scheduled one. We derive a deterministic error bound for the shrinkage and empirically evaluate its validity and tightness across acceleration regimes. At the same number of denoiser evaluations, RACER improves the strongest open-loop baseline across SD3.5-Large, FLUX.1-dev, Wan2.1-14B, and HunyuanVideo on DrawBench, VBench, and COCO. On SD3.5, we further show that RACER samples faster at equal quality. RACER generalizes across forecasting designs as well. For example, it recovers much of the quality lost on a Taylor base. These results show that reliable diffusion acceleration also depends on how forecasts are used. Code is available at https://github.com/LiZaiyuan0619/RACER

No Time Like the Present: Agentic Test-Time Training for LLM Agents

LLM agents often degrade over long episodes: as trajectories grow, they revisit explored states, repeat failed actions, and lose strategies that previously worked. Test-time training (TTT) offers a way to adapt model weights to the evolving task state, but existing LLM TTT methods largely adapt once to a fixed input. We study continuous TTT in multi-turn agent episodes, where each update changes the policy that generates later training text. This creates a self-training loop that helps when new trajectory information appears, but can amplify drift when the agent gets stuck and repeatedly trains on similar text. We find that update-text repetition distinguishes these regimes and introduce Agentic Test-Time Training (aTTT), a token-level reweighting method that downweights the loss on tokens appearing in repeated $n$-grams from prior updates while leaving novel tokens fully weighted. To run such updates inside live episodes, we build a concurrent serving system using vLLM's runtime LoRA API, limiting overhead to 1.9$\times$ the no-TTT cost. aTTT improves success by up to 5.0 points on ALFWorld and 4.9 points on SWE-bench Lite. The gains concentrate where models already have task competence but drift over long trajectories, suggesting that aTTT mainly preserves existing competence rather than teaching new abilities.

MeanVC 2: Robust Low-Latency Streaming Zero-Shot Voice Conversion

Streaming zero-shot voice conversion (VC) has become increasingly popular due to its potential for real-time applications. The recently proposed MeanVC achieves lightweight streaming zero-shot VC, but it has several limitations: its chunk-wise autoregressive denoising doubles the effective training sequence length, conversion quality degrades under small-chunk settings, and its timbre encoder directly relies on reference mel-spectrograms, making it sensitive to reference audio quality. To address these limitations we propose MeanVC 2. We introduce future-receptive chunking (FRC), which explicitly schedules past and future receptive fields across diffusion transformer decoder layers and removes clean-chunk teacher forcing. By incorporating bounded future context, FRC enables stable conversion with a 40 ms chunk size. We further introduce a universal timbre token encoder, which constructs a timbre representation from a global speaker embedding and retrieves fine-grained timbre cues via cross-attention, improving robustness to low-quality references and enhancing zero-shot speaker similarity. Experimental results show that MeanVC 2 significantly outperforms MeanVC, while reducing latency from 211 ms to 110 ms. Audio samples are publicly available. The source code will be publicly released.

SAGE: A Self-Evolving Agentic Graph-Memory Engine for Structure-Aware Associative Memory

Long-term memory is becoming a central bottleneck for language agents. Exsting RAG and GraphRAG systems largely treat memory graphs as static retrieval middleware, which limits their ability to recover complete evidence chains from partial cues, exploit reusable graph-structrual roles, and improve the memory itself through downstream feedback. We introduce SAGE, a Self-evolving Agentic Graph-memory Engine that models graph memory as a dynamic long-term memory substrate. SAGE couples two roles: a memory writer that incrementally constucts structured graph memory from interaction histories, and a Graph Foundation Model-based memory reader to perform retrieval and provide feedback to the memory writer. We provide rigorooous theoretical annalyses supporting the framework. Across multi-hop QA, open-domain retireval, domain-specific review QA, and long-term agent-memory benchmarks, SAGE improves evidence recovery, answer grounding, and retrieval efficiency: after two self-evolution rounds, it achieves the best average rank on multi-hop QA; in zero-shot open-domain transfer, it reaches 82.5/91.6 Recall@2/5 on NQ. Further results on LongMemEval and HaluMem show that traning and reader-writer feedback improve multiple long-term memory and hallucination-diagnostic metrics, suggesting that self-evolving, structure-aware graph memory is a promising foundation for robust long-horizon language agents.

Position: How can Graphs Help Large Language Models?

With the rapid advancement of large language models (LLMs), classic graph learning tasks have greatly benefited from LLMs, including improved encoding of textual features, more efficient construction of graphs from text, and enhanced reasoning over knowledge graphs. In this paper, we ask a complementary question: How can graphs help LLMs? We address this question from three perspectives: 1) graphs provide an up-to-date knowledge source that helps reduce LLM hallucinations, 2) graph-based prompting techniques-such as Chain-of-Thought (CoT), Tree-of-Thought (ToT), and Graph-of-Thought (GoT)-enhance LLM reasoning capabilities, and 3) integrating graphs into LLMs improves their understanding of structured data, expanding their applicability to domains such as e-commerce, code, and relational databases (RDBs). We further outlook some future directions including designing sparse LLM architectures based on graphs and brain-inspired memory systems.

Guardian-as-an-Advisor: Advancing Next-Generation Guardian Models for Trustworthy LLMs

Hard-gated safety checkers often over-refuse and misalign with a vendor's model spec; prevailing taxonomies also neglect robustness and honesty, yielding safer-on-paper yet less useful systems. This work introduces Guardian-as-an-Advisor (GaaA), a soft-gating pipeline where a guardian predicts a binary risk label plus a concise explanation and prepends this advice to the original query for re-inference, keeping the base model operating under its original spec. To support training and evaluation, GuardSet is constructed, a 208k+ multi-domain dataset unifying harmful and harmless cases with targeted robustness and honesty slices. GuardAdvisor is trained via SFT followed by RL to enforce label-explanation consistency. GuardAdvisor attains competitive detection accuracy while enabling the advisory workflow; when used to augment inputs, responses improve over unaugmented prompts. A latency study shows advisor inference uses below 5% of base-model compute and adds only 2-10% end-to-end overhead under realistic harmful-input rates. Overall, GaaA steers models to comply with the model spec, maintaining safety while reducing over-refusal.

Shorter, but Still Trustworthy? An Empirical Study of Chain-of-Thought Compression

Long chain-of-thought (Long-CoT) reasoning models have motivated a growing body of work on compressing reasoning traces to reduce inference cost, yet existing evaluations focus almost exclusively on task accuracy and token savings. Trustworthiness properties, whether acquired or reinforced through post-training, are encoded in the same parameter space that compression modifies. This means preserving accuracy does not, a priori, guarantee preserving trustworthiness. We conduct the first systematic empirical study of how CoT compression affects model trustworthiness, evaluating multiple models of different scales along three dimensions: safety, hallucination resistance, and multilingual robustness. Under controlled comparisons, we find that CoT compression frequently introduces trustworthiness regressions and that different methods exhibit markedly different degradation profiles across dimensions. To enable fair comparison across bases, we propose a normalized efficiency score for each dimension that reveals how naïve scalar metrics can obscure trustworthiness trade-offs. As an existence proof, we further introduce an alignment-aware DPO variant that reduces CoT length by 19.3\% on reasoning benchmarks with substantially smaller trustworthiness loss. Our findings suggest that CoT compression should be optimized not only for efficiency but also for trustworthiness, treating both as equally important design constraints.

Tencent Advertising Algorithm Challenge 2025: All-Modality Generative Recommendation

Generative recommender systems are rapidly emerging as a new paradigm for recommendation, where collaborative identifiers and/or multi-modal content are mapped into discrete token spaces and user behavior is modelled with autoregressive sequence models. Despite progress on multi-modal recommendation datasets, there is still a lack of public benchmarks that jointly offer large-scale, realistic and fully all-modality data designed specifically for generative recommendation (GR) in industrial advertising. To foster research in this direction, we organised the Tencent Advertising Algorithm Challenge 2025, a global competition built on top of two all-modality datasets for GR: TencentGR-1M and TencentGR-10M. Both datasets are constructed from real de-identified Tencent Ads logs and contain rich collaborative IDs and multi-modal representations extracted with state-of-the-art embedding models. The preliminary track (TencentGR-1M) provides 1 million user sequences with up to 100 interacted items each, where each interaction is labeled with exposure and click signals, while the final track (TencentGR-10M) scales this to 10 million users and explicitly distinguishes between click and conversion events at both the sequence and target level. This paper presents the task definition, data construction process, feature schema, baseline GR model, evaluation protocol, and key findings from top-ranked and award-winning solutions. Our datasets focus on multi-modal sequence generation in an advertising setting and introduce weighted evaluation for high-value conversion events. We release our datasets at https://huggingface.co/datasets/TAAC2025 and baseline implementations at https://github.com/TencentAdvertisingAlgorithmCompetition/baseline_2025 to enable future research on all-modality generative recommendation at an industrial scale. The official website is https://algo.qq.com/2025.

KAT-Coder-V2 Technical Report

We present KAT-Coder-V2, an agentic coding model developed by the KwaiKAT team at Kuaishou. KAT-Coder-V2 adopts a "Specialize-then-Unify" paradigm that decomposes agentic coding into five expert domains - SWE, WebCoding, Terminal, WebSearch, and General - each undergoing independent supervised fine-tuning and reinforcement learning, before being consolidated into a single model via on-policy distillation. We develop KwaiEnv, a modular infrastructure sustaining tens of thousands of concurrent sandbox instances, and scale RL training along task complexity, intent alignment, and scaffold generalization. We further propose MCLA for stabilizing MoE RL training and Tree Training for eliminating redundant computation over tree-structured trajectories with up to 6.2x speedup. KAT-Coder-V2 achieves 79.6% on SWE-bench Verified (vs. Claude Opus 4.6 at 80.8%), 88.7 on PinchBench (surpassing GLM-5 and MiniMax M2.7), ranks first across all three frontend aesthetics scenarios, and maintains strong generalist scores on Terminal-Bench Hard (46.8) and tau^2-Bench (93.9). Our model is publicly available at https://streamlake.com/product/kat-coder.