qwen3.5:4b
A small open-source large language model in the Qwen series with 4B parameters, suited to local deployment and tasks such as coding agents.
Timeline (last 90 days)
Under the old Agent framework, the qwen3.5:4b model repeatedly read the same file in task T3, claiming 'I read it but can't see the content,' and ultimately exhausted all six rounds and failed the task.
In the control experiment, the same 4B model exhausted all six rounds and failed under the old framework, while after the fix it completed in five rounds (pytest→read→fix→re-verify, correct).
In the request type judgment task, the ordinary qwen3:8b and qwen3.5:4b models achieved an accuracy of about 82%.
On "continuation of previous operation" type utterances, Qwen3.6-35B-A3B achieved an accuracy of 86.3%, while the 4B and 8B models achieved only 70.4%.
All Facts (4)
In the control experiment, the same 4B model exhausted all six rounds and failed under the old framework, while after the fix it completed in five rounds (pytest→read→fix→re-verify, correct).
50%UnverifiedIn the request type judgment task, the ordinary qwen3:8b and qwen3.5:4b models achieved an accuracy of about 82%.
50%UnverifiedOn "continuation of previous operation" type utterances, Qwen3.6-35B-A3B achieved an accuracy of 86.3%, while the 4B and 8B models achieved only 70.4%.
50%UnverifiedUnder the old Agent framework, the qwen3.5:4b model repeatedly read the same file in task T3, claiming 'I read it but can't see the content,' and ultimately exhausted all six rounds and failed the task.
50%