Archived
2026-09-23SWE-Serve: Benchmarking Agentic Engineering For Production Inference Serving
Jennifer Williams, Dave Farris, Jeff Farris, Jiantao Jiao
arXiv:2609.26777v1Today’s paper introduces SWE-Serve, a benchmark for a problem the field has struggled to measure well: can AI agents actually ship production inference features, not just hack together code that passes local tests? The authors build 53 repository-grounded tasks from real changes in the SGLang serving stack, covering model support, runtime behavior, APIs, and performance constraints. Each task is evaluated with hidden functional and regression tests, and where possible, end-to-end serving checks and calibrated performance gates. That matters because production inference work is full of cross-cutting details that ordinary software benchmarks miss. In their experiments, the best of 11 models and 31 effort settings reaches 75 percent pass at one, but the benchmark also exposes a big gap between “looks correct” and “works in production.” In particular, end-to-end tests reject many patches that would otherwise pass. SWE-Serve gives the community a much sharper way to track whether coding agents are becoming genuinely deployment-ready.
Also spotted that day
Flash-dLLM: IO-Aware KV Caching and Parallel Decoding for Fast, Memory-Efficient Diffusion LLMsA Decentralized Partially Observable Team Decision Methodology with Delayed Information SharingAgensh: Scaling Organizational Intelligence to 1,024 AgentsSpeakerMem-R1: Speaker-Centered Dual-Track Memory for Multi-Party DialogueCliffCompaction: Cost-Efficient Compaction for Long-Horizon Coding AgentsA2M: Trace-Optimized Agent Hijacking in the MCP EcosystemGrow the Harness, Not the Context: From Strategy-Free Scaffolds to Reusable Specialist AgentsType-Safe Is Not Error-Free: A Constrained Decision Head Follows the Option Name, Not the Rubric Bound to ItFleXray: Universal Clinical X-ray SegmentationEquivSVA: A Formally Verified Dataset of Behavioral Assertions Across Equivalent RTL ImplementationsMetrics Failure in LLM-Based Code Vulnerability Repair: An Empirical Study and a Change-Aware ScreenAutomatic depth-based local center clustering via $β$-integrated local depth and adaptive groupingDiffusion-Induced Spatial Attention Overlapping Community DetectionDoes AI Save Time on Product Design? A Randomized Controlled Experiment of AI Prompt-to-Design Workflows
Previous daily papers
2026-09-26PoEM: Predicting RL Outcomes from Existing Policies2026-09-26LLM Agents Can Easily Tamper With Their Own Traces2026-09-26Instrumental Monitor Evasion Emerges Under Ordinary Task Pressure2026-09-25TrackEverything: Long Horizon Dense Tracking via De-Duplicating 3D Scene Representations2026-09-24Frozen Flows Forget: Diagnosing and Restoring Lost Motion in a Latent-flow World Model2026-09-24StudentBench: AI and human tutoring yield equivalent GRE learning gains2026-09-24Agent-Editing World Model: Rethinking World Modeling for LLM Agents2026-09-23Grow the Harness, Not the Context: From Strategy-Free Scaffolds to Reusable Specialist Agents2026-09-23FleXray: Universal Clinical X-ray Segmentation2026-09-22GameHorizon Suite: Multi-Horizon Data and Evaluation in Gameplay2026-09-22Emergent Collusion in Long-Horizon LLM Agent Interaction2026-09-22WorldCrafter: Consistent Video World Model with Implicit 3D-aware Memory2026-09-21Value-Sensitive Delegation in Everyday AI Agent Use: Evidence from OpenClaw2026-09-21Designer-RSI: Evolving Procedural Memory from User Traffic for Agentic Graphic Design2026-09-21CodeMidas: Scaling Agentic Coding RL Environments from Code Itself2026-09-20RetireOPD: Self-Retiring On-Policy Distillation for Agentic Reinforcement Learning2026-09-20Quantifying Overclaiming Propensity in Frontier LLM Agents2026-09-20JEPA-Anything: Learning Predictive Models across Different Worlds2026-09-19An Empirical Study of Harness Design for Coding Agents2026-09-17rMuscle: Robotic Muscle Memory for Efficient Vision-Language-Action Model Inference2026-09-17How Model Growth, Recursion, and Boundary Operators Influence Scaling Exponents2026-09-17Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluations2026-09-16JustFit: 200K-Token LLM Serving on a 24 GiB Laptop with Just-in-Time State Management2026-09-16Agentic Societies Need a Social Harness2026-09-16Coupled Calibration and Learning: Mitigating Teacher Bias in LLM Distillation without Target-Domain Reward Feedback2026-09-15Corrupt Plans, Clean Traces: Evading Chain-of-Thought Monitoring with Plan Injection2026-09-14A Ranking Approach for Measuring Calibration2026-09-12GPU-CFR: 80x Faster Counterfactual Regret Minimization by Compiling the Game to Static Dataflow and CUDA Graph Replay2026-09-12CausalArena: Benchmarking Causal Discovery in the Foundation Model Era2026-09-11Data Scarcity and Model Sparsity: Mixtures-of-Experts Overfit More to Repeated Data2026-09-11General Quantification of Covariate and Concept Shifts2026-09-10Show-Harness: Just a VLM Agent Can Play Robots2026-09-10JarvisGUI: Towards Cross-Device GUI Agents with Dynamic Task Composition2026-09-10Building Multilingual Bridges: Data Mixing as the Pillar of Generalization for In-Language Reasoning2026-09-09ExecCritic: Learn to Test, Test to Improve for Coding Agents2026-09-09Learning Length-Extrapolatable Recurrent Models2026-09-09TANGO: Humanoid Navigation in Cluttered Environments with a Whole-Body Vision-Language-Action Model2026-09-08WearableQA: A Benchmark for Health Reasoning over Real-World Wearable Data2026-09-08UniMate: One Unified Model to Animate Diverse Skeletons2026-09-08CUA-Universe: A Scalable and Dynamic Environment for Hybrid GUI+CLI Agents2026-09-06Compile by Training: Turning Natural-Language Specifications into Local Neural Functions2026-09-06Rethinking On-Policy Distillation of Large Language Models II: One Training Example2026-09-06SWE-Gate: Passing Functional Tests Is Not Enough for Software Engineering Agents2026-09-05From Deceptive Outputs to Deceptive Mechanisms: A Causal Framework for Language-Model Deception Research2026-09-05Last Translation Benchmark2026-09-03A Common Measure of Communication for Speech Brain-Computer Interfaces2026-09-03Discriminative World Models for Web Agents2026-09-03Post-Training Language Models for Gold-Medal Performance in Coding Competitions2026-09-02Mechanism Design for Alignment and Control2026-09-02Beyond Scores: Understanding LLM-as-a-Judge Mechanisms in Summarization Evaluation2026-09-02The Rise of Verbal Reinforcement Learning2026-09-01Stress-Testing Efficient Responsible-AI Evaluation: When Compute Savings Change Benchmark Conclusions2026-09-01Aspire: Can Models Self-Evolve from Vague Goals?2026-09-01PaperGym: Rubric-Centered Evolution for Research-Plan Generation2026-08-31DARTS: Decoder-Aware Representation Tuning via Surgery for Model Merging2026-08-31Video Generative Models as Geometry Learner2026-08-31When Robots Mishear Us: Mapping the Safety Risks of Voice-Controlled Embodied AI2026-08-30CritICL: Inference-Time Weak-to-Strong Generalization from Small Language Model Failure Modes2026-08-30Boosting LLM Exploration via Weak-Model Guidance in RLVR