CiteVQA requires models to cite specific document regions with bounding boxes alongside answers and finds that even the strongest MLLMs frequently cite the wrong region, with top SAA scores of only 76.0 for closed models and 22.5 for open-source ones.
hub Mixed citations
GLM-5: from Vibe Coding to Agentic Engineering
Mixed citation behavior. Most common role is background (68%).
79
Pith papers citing it
Background
68% of classified citations
abstract
We present GLM-5, a next-generation foundation model designed to transition the paradigm of vibe coding to agentic engineering. Building upon the agentic, reasoning, and coding (ARC) capabilities of its predecessor, GLM-5 adopts DSA to significantly reduce training and inference costs while maintaining long-context fidelity. To advance model alignment and autonomy, we implement a new asynchronous reinforcement learning infrastructure that drastically improves post-training efficiency by decoupling generation from training. Furthermore, we propose novel asynchronous agent RL algorithms that further improve RL quality, enabling the model to learn from complex, long-horizon interactions more effectively. Through these innovations, GLM-5 achieves state-of-the-art performance on major open benchmarks. Most critically, GLM-5 demonstrates unprecedented capability in real-world coding tasks, surpassing previous baselines in handling end-to-end software engineering challenges. Code, models, and more information are available at https://github.com/zai-org/GLM-5.
hub tools
citation-role summary
background 22
baseline 6
dataset 2
method 2
other 2
citation-polarity summary
claims ledger
- abstract We present GLM-5, a next-generation foundation model designed to transition the paradigm of vibe coding to agentic engineering. Building upon the agentic, reasoning, and coding (ARC) capabilities of its predecessor, GLM-5 adopts DSA to significantly reduce training and inference costs while maintaining long-context fidelity. To advance model alignment and autonomy, we implement a new asynchronous reinforcement learning infrastructure that drastically improves post-training efficiency by decoupling generation from training. Furthermore, we propose novel asynchronous agent RL algorithms that fur
co-cited works
years
2026 79representative citing papers
A new native-runtime benchmark reveals that current frontier AI agents succeed on at most 62 percent of realistic long-horizon CLI tasks.
Agent-ValueBench is the first dedicated benchmark for agent values, showing they diverge from LLM values, form a homogeneous 'Value Tide' across models, and bend under harnesses and skill steering.
SignSGD provably beats SGD by a factor of d under sparse noise via matched ℓ1-norm upper and lower bounds, with an equivalent result for Muon on matrices, and this predicts faster GPT-2 pretraining.
OccuBench is a new benchmark for AI agents on real-world occupational tasks via LLM-driven simulators, showing no model dominates all industries, implicit faults are hardest, and larger models with more reasoning perform better.
AMUSE is a new optimizer integrating Muon orthogonalization with Schedule-Free averaging via adaptive interpolation for schedule-free anytime training that improves Pareto frontiers on vision and LLM tasks.
Terminal-World is a skill-based synthesis pipeline that generates 5,723 training environments and produces Terminal-World-32B which outperforms baselines on Terminal-Bench 2.0 using only 1.2% of the data.
WebGameBench is a new benchmark that evaluates coding agents on building browser-native games from frozen specifications, with runtime browser evaluation showing best agents reach 76.9% usable rate but only 20.2% excellent rate.
Decoupling prefix source from token-level KL direction in autoregressive sequence KL yields four objectives unifying SFT, DAgger, offline RL and OPD, with KL mixing and entropy-gated curriculum improving math reasoning accuracy and shortening responses.
AssayBench is a new gene-ranking benchmark for phenotypic CRISPR screens that shows zero-shot generalist LLMs outperform both biology-specific LLMs and trainable baselines on adjusted nDCG.
StereoTales shows that all tested LLMs emit harmful stereotypes in open-ended stories, with associations adapting to prompt language and targeting locally salient groups rather than transferring uniformly across languages.
LLM proofs for hard math problems show large differences in quality metrics like conciseness and cognitive simplicity that correctness-only tests miss, along with trade-offs between quality and correctness.
OPHSD uses harness-augmented models as teachers to distill reasoning capabilities into base LLMs, yielding strong standalone performance on classification and math tasks.
vOPD stabilizes on-policy distillation gradients by subtracting a closed-form per-token negative reverse KL baseline as a detached control variate, preserving unbiasedness while lowering variance and matching expensive full-vocabulary methods.
MISA routes to a small subset of indexer heads via block statistics, matching full DSA performance on LongBench with 4-8x fewer heads and 3.82x speedup while recovering over 92% of selected tokens.
TEBench is a new project-level benchmark for test evolution showing coding agents achieve only 45-49% F1 on identifying tests needing changes, with stale tests hardest due to reliance on execution failures.
MolViBench is the first benchmark designed to evaluate LLMs on generating executable programs for molecular tasks in drug discovery.
A malicious relay can strategically rewrite aligned LLM outputs in BYOK agent architectures to achieve up to 99.1% attack success on benchmarks like AgentDojo and ASB.
Self-play between LLMs for problem authoring and solving, scored via Rasch modeling, shows that authoring and solving skills are partially decoupled and that the benchmark difficulty evolves with new models.
MultiLogBench shows that LLM performance on automated logging varies substantially across programming languages, demonstrating that single-language evidence is insufficient for general claims about model behavior or tool design.
BARD bridges autoregressive and diffusion VLMs with progressive block merging plus stage-wise intra-diffusion distillation, delivering 3x speedup and new SOTA on open dVLMs using under 4.4M data points.
SkillMOO applies LLM-proposed edits and NSGA-II Pareto optimization to skill bundles for SE agents, ranking top in pass rate on most SkillsBench tasks while cutting costs up to 31.7%.
ClawBench is a benchmark of 153 live-web tasks where AI agents achieve low success rates, e.g. 33.3% for Claude Sonnet 4.6.
AgentHazard benchmark shows computer-use agents remain highly vulnerable, with attack success rates reaching 73.63% on models like Qwen3-Coder powering Claude Code.
citing papers explorer
-
CiteVQA: Benchmarking Evidence Attribution for Trustworthy Document Intelligence
CiteVQA requires models to cite specific document regions with bounding boxes alongside answers and finds that even the strongest MLLMs frequently cite the wrong region, with top SAA scores of only 76.0 for closed models and 22.5 for open-source ones.
-
WildClawBench: A Benchmark for Real-World, Long-Horizon Agent Evaluation
A new native-runtime benchmark reveals that current frontier AI agents succeed on at most 62 percent of realistic long-horizon CLI tasks.
-
Agent-ValueBench: A Comprehensive Benchmark for Evaluating Agent Values
Agent-ValueBench is the first dedicated benchmark for agent values, showing they diverge from LLM values, form a homogeneous 'Value Tide' across models, and bend under harnesses and skill steering.
-
When and Why SignSGD Outperforms SGD: A Theoretical Study Based on $\ell_1$-norm Lower Bounds
SignSGD provably beats SGD by a factor of d under sparse noise via matched ℓ1-norm upper and lower bounds, with an equivalent result for Muon on matrices, and this predicts faster GPT-2 pretraining.
-
OccuBench: Evaluating AI Agents on Real-World Professional Tasks via Language Environment Simulation
OccuBench is a new benchmark for AI agents on real-world occupational tasks via LLM-driven simulators, showing no model dominates all industries, implicit faults are hardest, and larger models with more reasoning perform better.
-
AMUSE: Anytime Muon with Stable Gradient Evaluation
AMUSE is a new optimizer integrating Muon orthogonalization with Schedule-Free averaging via adaptive interpolation for schedule-free anytime training that improves Pareto frontiers on vision and LLM tasks.
-
Terminal-World: Scaling Terminal-Agent Environments via Agent Skills
Terminal-World is a skill-based synthesis pipeline that generates 5,723 training environments and produces Terminal-World-32B which outperforms baselines on Terminal-Bench 2.0 using only 1.2% of the data.
-
WebGameBench: Requirement-to-Application Evaluation for Coding Agents via Browser-Native Games
WebGameBench is a new benchmark that evaluates coding agents on building browser-native games from frozen specifications, with runtime browser evaluation showing best agents reach 76.9% usable rate but only 20.2% excellent rate.
-
Decoupling KL and Trajectories: A Unified Perspective for SFT, DAgger, Offline RL, and OPD in LLM Distillation
Decoupling prefix source from token-level KL direction in autoregressive sequence KL yields four objectives unifying SFT, DAgger, offline RL and OPD, with KL mixing and entropy-gated curriculum improving math reasoning accuracy and shortening responses.
-
AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents
AssayBench is a new gene-ranking benchmark for phenotypic CRISPR screens that shows zero-shot generalist LLMs outperform both biology-specific LLMs and trainable baselines on adjusted nDCG.
-
StereoTales: A Multilingual Framework for Open-Ended Stereotype Discovery in LLMs
StereoTales shows that all tested LLMs emit harmful stereotypes in open-ended stories, with associations adapting to prompt language and targeting locally salient groups rather than transferring uniformly across languages.
-
Not All Proofs Are Equal: Evaluating LLM Proof Quality Beyond Correctness
LLM proofs for hard math problems show large differences in quality metrics like conciseness and cognitive simplicity that correctness-only tests miss, along with trade-offs between quality and correctness.
-
Training with Harnesses: On-Policy Harness Self-Distillation for Complex Reasoning
OPHSD uses harness-augmented models as teachers to distill reasoning capabilities into base LLMs, yielding strong standalone performance on classification and math tasks.
-
KL for a KL: On-Policy Distillation with Control Variate Baseline
vOPD stabilizes on-policy distillation gradients by subtracting a closed-form per-token negative reverse KL baseline as a detached control variate, preserving unbiasedness while lowering variance and matching expensive full-vocabulary methods.
-
MISA: Mixture of Indexer Sparse Attention for Long-Context LLM Inference
MISA routes to a small subset of indexer heads via block statistics, matching full DSA performance on LongBench with 4-8x fewer heads and 3.82x speedup while recovering over 92% of selected tokens.
-
Breaking, Stale, or Missing? Benchmarking Coding Agents on Project-Level Test Evolution
TEBench is a new project-level benchmark for test evolution showing coding agents achieve only 45-49% F1 on identifying tests needing changes, with stale tests hardest due to reliance on execution failures.
-
MolViBench: Evaluating LLMs on Molecular Vibe Coding
MolViBench is the first benchmark designed to evaluate LLMs on generating executable programs for molecular tasks in drug discovery.
-
When Alignment Isn't Enough: Response-Path Attacks on LLM Agents
A malicious relay can strategically rewrite aligned LLM outputs in BYOK agent architectures to achieve up to 99.1% attack success on benchmarks like AgentDojo and ASB.
-
MathDuels: Evaluating LLMs as Problem Posers and Solvers
Self-play between LLMs for problem authoring and solving, scored via Rasch modeling, shows that authoring and solving skills are partially decoupled and that the benchmark difficulty evolves with new models.
-
Single-Language Evidence Is Insufficient for Automated Logging: A Multilingual Benchmark and Empirical Study with LLMs
MultiLogBench shows that LLM performance on automated logging varies substantially across programming languages, demonstrating that single-language evidence is insufficient for general claims about model behavior or tool design.
-
BARD: Bridging AutoRegressive and Diffusion Vision-Language Models Via Highly Efficient Progressive Block Merging and Stage-Wise Distillation
BARD bridges autoregressive and diffusion VLMs with progressive block merging plus stage-wise intra-diffusion distillation, delivering 3x speedup and new SOTA on open dVLMs using under 4.4M data points.
-
SkillMOO: Multi-Objective Optimization of Agent Skills for Software Engineering
SkillMOO applies LLM-proposed edits and NSGA-II Pareto optimization to skill bundles for SE agents, ranking top in pass rate on most SkillsBench tasks while cutting costs up to 31.7%.
-
ClawBench: Can AI Agents Complete Everyday Online Tasks?
ClawBench is a benchmark of 153 live-web tasks where AI agents achieve low success rates, e.g. 33.3% for Claude Sonnet 4.6.
-
AgentHazard: A Benchmark for Evaluating Harmful Behavior in Computer-Use Agents
AgentHazard benchmark shows computer-use agents remain highly vulnerable, with attack success rates reaching 73.63% on models like Qwen3-Coder powering Claude Code.
-
DWDP: Distributed Weight Data Parallelism for High-Performance LLM Inference on NVL72
DWDP distributes MoE weights across GPUs for independent execution without collective synchronization, improving output TPS/GPU by 8.8 percent on GB200 NVL72 for DeepSeek-R1 under 8K input and 1K output lengths.
-
PERMA: Benchmarking Personalized Memory Agents via Event-Driven Preference and Realistic Task Environments
PERMA is a new benchmark using temporally ordered events, text variability, and linguistic alignment to evaluate LLM memory agents on persona consistency beyond simple retrieval.
-
OnePred: Next-Query Prediction via Recursive Intent Memory in Multi-Turn Conversations
OnePred maintains a recursively updated intent memory and uses two-stage RL to predict next queries, cutting token use by up to 22x while outperforming baselines on a new NQP-Bench dataset.
-
Towards Direct Evaluation of Harness Optimizers via Priority Ranking
Priority ranking offers a low-cost direct evaluation for harness optimizers that correlates with their real multi-step optimization performance, supported by the Shor dataset of 182 scenarios.
-
Benchmarking Autonomous Agents against Temporal, Spatial, and Semantic Evasions
A3S-Bench evaluates LLM agents against temporal, spatial, and semantic evasions, raising average risk trigger rates from 28.3% to 52.6% across 2,254 trajectories and 20 scenarios.
-
From Patches to Trajectories: Privileged Process Supervision for Software-Engineering Agents
P2T distills reference patches into a latent process graph and uses it to select shortest effective trajectory segments from teacher rollouts, yielding up to 10.8 point Pass@1 gains on SWE-bench Verified with 15% lower inference cost using only 1.8k instances.
-
Post-Trained MoE Can Skip Half Experts via Self-Distillation
ZEDA injects zero-output experts and uses two-stage self-distillation to adapt post-trained MoE models into dynamic ones that skip over half the experts, yielding 1.2x inference speedup with small accuracy drops.
-
CHI-Bench: Can AI Agents Automate End-to-End, Long-Horizon, Policy-Rich Healthcare Workflows?
CHI-Bench shows current AI agents achieve at most 28% success on long-horizon healthcare workflows that require dense policy adherence, multi-role handoffs, and multi-turn interactions.
-
Argus: Evidence Assembly for Scalable Deep Research Agents
Argus coordinates a Navigator and multiple Searchers via an evidence graph for deep research, reporting average gains of 5.5 points with one Searcher and 12.7 points with eight parallel Searchers across eight benchmarks, reaching 86.2 on BrowseComp with 64 Searchers.
-
Harnesses for Inference-Time Alignment over Execution Trajectories
Partial harnesses for LLM agents, specifying only initial execution steps, achieve higher pass rates than fully decomposed workflows, as analyzed through trajectory alignment and validated in synthetic and terminal benchmarks.
-
Exploiting LLM Agent Supply Chains via Payload-less Skills
Semantic Compliance Hijacking lets attackers hijack LLM agents by disguising malicious instructions as compliance rules in skills, reaching up to 77.67% success on confidentiality breaches and 67.33% on RCE while evading all tested scanners.
-
SWE-Cycle: Benchmarking Code Agents across the Complete Issue Resolution Cycle
SWE-Cycle benchmark shows sharp drops in code agent success rates from isolated tasks to full autonomous issue resolution, highlighting cross-phase dependency issues.
-
ToolCUA: Towards Optimal GUI-Tool Path Orchestration for Computer Use Agents
ToolCUA introduces a trajectory scaling pipeline and staged RL to optimize GUI-tool switching, reaching 46.85% accuracy on OSWorld-MCP for a 66% relative gain over baseline.
-
Missing Old Logits in Asynchronous Agentic RL: Semantic Mismatch and Repair Methods for Off-Policy Correction
Missing old logits in async agentic RL entangle discrepancy and staleness terms in PPO off-policy correction; exact acquisition methods and revised PPO-EWMA restore decoupled updates with reported gains in speed and performance.
-
ProteinOPD: Towards Effective and Efficient Preference Alignment for Protein Design
ProteinOPD uses token-level on-policy distillation from multiple preference-specific teacher models into a shared student to balance competing objectives in protein design, delivering gains on targets without losing designability and an 8x speedup over RL baselines.
-
LoopTrap: Termination Poisoning Attacks on LLM Agents
LoopTrap is an automated red-teaming framework that crafts termination-poisoning prompts to amplify LLM agent steps by 3.57x on average (up to 25x) across 8 agents.
-
Evaluation Awareness in Language Models Has Limited Effect on Behaviour
Verbalised evaluation awareness in large reasoning models has only small effects on their outputs across safety and alignment tests.
-
Uni-OPD: Unifying On-Policy Distillation with a Dual-Perspective Recipe
Uni-OPD unifies on-policy distillation across LLMs and MLLMs with dual-perspective strategies that promote student exploration and enforce order-consistent teacher supervision based on outcome rewards.
-
Affordance Agent Harness: Verification-Gated Skill Orchestration
Affordance Agent Harness is a verification-gated orchestration system that unifies skills via an evidence store, episodic memory priors, an adaptive router, and a self-consistency verifier to improve accuracy-cost tradeoffs in open-world affordance grounding.
-
Making Every Verified Token Count: Adaptive Verification for MoE Speculative Decoding
EVICT adaptively truncates draft trees in MoE speculative decoding by combining drafter signals with profiled costs to retain only cost-effective prefixes, delivering up to 2.35x speedup over autoregressive decoding.
-
Co-Evolving Policy Distillation
CoPD integrates multiple expert capabilities by running parallel RLVR training with bidirectional online policy distillation among experts, outperforming mixed RLVR and sequential OPD while surpassing domain-specific experts on text-image-video reasoning.
-
MAIC-UI: Making Interactive Courseware with Generative UI
MAIC-UI provides a zero-code authoring system for generating and iteratively editing interactive courseware from educational materials via structured analysis and incremental generation, with lab and classroom evaluations showing usability gains and learning improvements.
-
QuantClaw: Precision Where It Matters for OpenClaw
QuantClaw dynamically routes precision in agent workflows to cut cost by up to 21.4% and latency by 15.7% while keeping or improving task performance.
-
Temporally Extended Mixture-of-Experts Models
Temporally extended MoE layers using the option-critic framework with deliberation costs cut switching rates below 5% while retaining most capability on MATH, MMLU, and MMMLU.
-
AeSlides: Incentivizing Aesthetic Layout in LLM-Based Slide Generation via Verifiable Rewards
AeSlides is a GRPO-based RL framework that uses verifiable aesthetic metrics to optimize LLM slide generation, achieving large gains in layout quality metrics and human scores with only 5K prompts.
-
ClawEnvKit: Automatic Environment Generation for Claw-Like Agents
ClawEnvKit automates generation of diverse verified environments for claw-like agents from natural language, producing the Auto-ClawEval benchmark of 1,040 environments that matches human-curated quality at 13,800x lower cost.