The paper investigates why large‑language‑model coding agents rarely request a second chunk of tool output, focusing on the precision‑at‑1 rate ($p_1$) of the gold item appearing first in the first chunk. In a benchmark of 500 software‑engineering tasks, the authors compare six value functions and find that increasing $p_1$ does not systematically improve downstream accuracy; the agent can recover the correct answer from any position within the chunk. Adding file‑metadata signals to a keyword scorer actually reduces $p_1$, while a parameter‑free keyword scorer improves $p_1$ but still fails to boost overall accuracy.
By Tatiana Petrova, Andrei Mazniak, Radu State
APEX-EM is a non‑parametric experience memory that stores full procedural‑episodic traces in a typed Procedural Knowledge Graph and retrieves them via semantic search, structural‑signature matching, and graph traversal. It uses a Plan‑Retrieve‑Generate‑Iterate‑Ingest workflow to produce, quality‑gate, and commit experiences, indexing both successes and failures so the agent learns what to reuse and what to avoid. Evaluations on five benchmarks with a shared GPT‑4o backbone show significant performance gains, such as +7.6 pp on BigCodeBench transfer and +1.4 pp on Lifelong Agent Bench, demonstrating that the memory adds to model capability rather than replacing it.
By Pratyay Banerjee, Masud Moshtaghi, Ankit Chadha
The paper investigates how scaling a team of small language‑model agents affects performance across different orchestration architectures. By testing eight architectures on five short‑answer benchmarks and an executable‑code benchmark, it finds that team scaling yields large gains on arithmetic word‑problem tasks but only modest improvements on multiple‑choice and code generation tasks, with no single architecture dominating all tasks. The authors explain these patterns using a generate‑transform decomposition that separates coverage and transformation effects, showing that arithmetic tasks benefit from both coverage and critic‑guided transformation, while other tasks are limited by saturation or poor conversion.
By Blaz Bertalanic, Carolina Fortuna
Mingbird is a local‑first agent harness designed for small open‑weight language models (2–9 B) that run on ordinary laptops. It introduces ten mechanisms—such as a byte‑level net‑zero prefill budget, a finish gate that re‑reads the task before accepting completion, and signature‑level loop detection—to address common failure modes that arise from the harness rather than the model itself. In controlled experiments on the LRAB benchmark and the $ au^2$‑bench, Mingbird achieves higher overall scores (0.886 and 0.856 respectively) compared to other harnesses, and its ablation studies show that each mechanism contributes measurable performance gains.
By Hao Wang, Ting Huang
arXiv:2609.23886v1 Announce Type: new
Abstract: Software delegates more of its branches to models every year: which queue a ticket enters, whether a command is safe to run, whether a claim clears wit...
By Zehua Cheng, Wei Dai, Jiahao Sun
arXiv:2609.11987v1 Announce Type: cross
Abstract: An agentic coding system couples a language model to a harness: the tools, prompts and control flow that turn a chat model into an autonomous softwar...
By Mohsen Arjmandi
arXiv:2610.02267v1 Announce Type: new
Abstract: Agent harnesses make many small, typed decisions per task: which model to call, which tool to use, whether retrieved text is relevant, whether an input...
By Jiawei Li
The paper introduces GVS5H, a ledger‑based zero‑shot self‑orchestration technique that allows fresh instances of a single model to decompose coding problems and coordinate via a shared file system. On the 100 most recent hard LiveCodeBench tasks, the method improves performance by up to 25.6 points across eleven open and closed‑weight models, enabling cheaper models to reach frontier‑level accuracy. While the gains are not universal—some models see no improvement or a decline—analysis suggests that decomposition and persistent context are key contributors to the observed performance boost.
By Victor Gao (Sang Won), Vida Khosrowshahi (Sang Won), Ali Khosrowshahi (Sang Won), Xihao Sun (Sang Won), Juhyun Lee (Sang Won), Simon (Sang Won), Lee
The paper introduces an open midtraining recipe that primes base models for agentic post‑training, filling a gap between pretraining and full fine‑tuning. It trains on 256K context with hundreds of billions of tokens drawn from natural software‑engineering data and a small share of agentic traces. Applying this midtraining to Nemotron 3.5 Lightning and Super models yields measurable gains on SWE‑bench and DeepSWE benchmarks after identical full SFT, and improves robustness across agent harnesses.
By Soumye Singhal, Joseph Jennings, Ka Shun Shum, Alexander Bukharin, Khushi Bhardwaj, Zirui Liu, Venkat Srinivasan, Jonathan Li, Yash Jain, Kezhi Kong, Markus Kliegl, Olivier Delalleau, Sanjeev Satheesh, Mostofa Patwary
Prefill‑only decision models evaluate every candidate in a menu in a single forward pass, avoiding decoding and reducing cost by one to two orders of magnitude compared to generative language models. The paper demonstrates that when only the candidate menu changes, the model’s post‑intervention accuracy can be predicted solely from the cached first‑pass distribution using a simple estimator that renormalizes and selects the argmax, without any labels or second pass. Across seven model families, ten datasets, and two task types, this menu‑only intervention prediction is within 4.2 points of actual accuracy, and in one family it is exact, whereas a probability‑level variant fails by 21 points, indicating the property resides in ranking rather than calibrated probabilities.
By Ran Li, Lei Chen
The paper introduces Galahad, a memory layer that stores a transformer language model’s key‑value state for blocks of text, allowing subsequent requests to reuse previously computed attention rather than recomputing it. On seven real‑world datasets, 98.7% of prompt tokens were already read, and with Galahad the model could attend to an entire 97,000‑token corpus, achieving 98–100% recall on a 100‑fact test while reducing inference time and energy consumption dramatically. The approach was validated across 30 models and all runtimes, demonstrating that stateful inference can replace stateless serving without loss of accuracy.
By Sietse Schelpe
The paper argues that post‑training pruning of large language models should use a calibration set that covers multiple capability domains rather than a single source. A balanced four‑way mixture of sources consistently outperforms any single source across sparsity levels, model scales, and held‑out benchmarks. When aligned corpora are unavailable, the IGSP method approximates this mixture using only the base model and an evaluation taxonomy, achieving significant gains over prior self‑generation approaches.
By Hu Xu, Zhaolong Xing, Congcong Liu, Jiaxing Wang, Zhida Jiang, Junshi Huang, Zhen Chen, Jianfeng Xu