Tag
Microsoft releases FrogNano-4B-2609, a compact repository-level coding agent derived from Qwen3.5-4B that is post-trained exclusively with reinforcement learning on ~1,500 synthetic SWE tasks via the TaskPilot pipeline and the lightweight five-tool Leaf harness, with no distillation from stronger models.
NVIDIA proposes HORIZON, a self-evolving agent framework that treats hardware design as repository-level code evolution, achieving 100% benchmark completion across several hardware design suites.
RepoExec is a new benchmark for evaluating repository-level code generation focused on executability, functional correctness via automated test generation, and cross-file dependency integration. Experiments reveal that pretrained LLMs beat instruction-tuned ones on correctness, while fine-tuned models better leverage provided code dependencies.