arxiv arXiv cs.CL · just now · research

JIT-Agent enables on-the-fly harness evolution to boost agentic LLM performance

from English

The authors present JIT-Agent, a harness intelligence model trained to synthesize task-adaptive agent harnesses for arbitrary off-the-shelf agentic LLMs. This approach formalizes the agent harness as a composable artifact governed by a fixed four-module protocol, allowing the system to customize and repair harnesses on the fly.

JIT-Agent distills performance signals from an expanding archive of prior configurations to self-evolve its strategies.
DeepSeek-V4-Flash surpasses GPT-5.6 on DeepSearchQA (+9.1) and OdysseyBench (+4.3) when equipped with JIT-Agent.
The GLM-5.2 model gains up to +20.2 points in performance improvements.
Generated harnesses are competitive with mature agent runtimes like OpenCode and Claude Code.
The method consistently improves multi-scale model families including DeepSeek V4, Mimo-V2.5, and Qwen3.6.

JIT-Agent establishes harness intelligence as a trainable, transferable, and compounding dimension of agent capability that is orthogonal to model scaling.

Importance 2/3 New harness with differentiators arXiv cs.CL DeepSeek AI agents Research paper

Read original

arxiv arXiv cs.CL · 9d ago · 4 views

Mint-Agent introduces finance-native agentic foundation models

Researchers present Mint-Agent, a family of finance-native agentic models designed to combine reliable execution with long-horizon research capabilities. The system is built on three pillars: a data engine for specialized tasks, the MintHarness environment for auditable evidence trails, and a training recipe combining SFT, OPD, and RLVR.

arxiv arXiv cs.LG · 13d ago · 25 views

Faraday 27B outperforms Claude Opus 4.8 and GPT-5.5 on paper replication tasks

Researchers have developed Replica, a scalable task space for paper replication, to address the need for reliable scientific knowledge. They post-trained Faraday, a 27B-parameter "AI Scientist" agent that uses coding agents as tools, to leverage this environment.

arxiv arXiv cs.AI · 13d ago · 21 views

Faraday 27B surpasses Claude Opus 4.8 and GPT-5.5 on paper replication tasks

Researchers have developed Replica, a scalable task space for paper replication, and introduced an auto-generated rubric-based judge to provide low-noise reward signals that align with human assessment.

arxiv arXiv cs.CL · 53d ago · 8 views

AutoMem automates learning of memory as a cognitive skill for LLMs

Researchers introduce AutoMem, a framework that treats memory management in large language models as a trainable skill rather than a static component. The system promotes file-system operations to first-class memory actions, allowing the model to autonomously decide how to encode, retrieve, and organize knowledge.

arxiv arXiv cs.CL · 58d ago · 7 views

Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent

Researchers introduce Agents-A1, a 35B Mixture-of-Experts model that achieves performance comparable to trillion-parameter models by scaling the agent horizon rather than parameter count. The approach focuses on extending long-horizon trajectories and unifying heterogeneous agent abilities through a specialized training infrastructure.