Can we get more details of qwen3-coder & SWE-Bench?

[FFengIll]

I am glad to see Qwen3-Coder win a high score in swe-bench, but I do not find any details (logs and trajectories) like other agent challengers (e.g. swe-bench, openhands). Can we get more details?

[cyente]

plz wait for our technique report

[ytxmobile98]

@cyente Hi, I am also new to the Qwen3-Coder model. I am wondering if you might also publicize the code for running SWE-bench against the Qwen3-Coder model? Thanks.

[Chen-GX]

@cyente Hi, I'm also wondering about the specific prompt or tool calling format for Qwen3-Coder model in Swe-bench. I've tested Qwen3-Coder in OpenHands scafflod with CodeXML format. I noticed some format error in Qwen3-Coder.

For example:

• The correct tool calling format in OpenHands is: <function=example_function_name> <parameter=example_parameter_1>value_1 <parameter=example_parameter_2>

• but the response in Qwen3-Coder may be <parameter=command=create</parameter>, which is a frequent issue.