Clarification on SWE-Bench Verified Evaluation Setup: Agent Framework and Max Turns

[Hellozaq]

Hello Qwen Team,
Thank you for releasing the impressive Qwen3.5 series.
I am writing to request some clarification regarding the specific evaluation settings used for the reported SWE-Bench Verified scores in the blog post. While the results are cited, the exact configuration of the agent harness is not fully detailed.
Could you please confirm the following details:
Agent Framework: Which specific agent framework or harness was used for the official SWE-Bench Verified evaluation? (e.g., Was it the standard SWE-Agent, OpenHands, or a custom internal framework?)
Maximum Conversation Turns: What was the maximum number of interaction turns (max_turns) allowed per instance during the evaluation? (e.g., 100, 500?)
Thank you for your time and for your continued contributions to the open-source community.