How well does this work with local models like https://github.com/receptron/laya (open source Jev) and vision capable models like Qwen 3.8 Flash Next, or even Qwen 3.8 27B on xhigh thinking?
How well does this work with local models like https://github.com/receptron/laya (open source Jev) and vision capable models like Qwen 3.8 Flash Next, or even Qwen 3.8 27B on xhigh thinking?
Didn't try this with laya, but just tested it with Qwen 3.8 27B via OpenRouter.
Tried twice, one starting directly in Agent mode, and another one starting first in Plan mode and then Agent.
Conclusions: very slow in general, starting in Plan mode gives way better and faster results, but still with defects.
See: - Notes: https://github.com/anteloc/ldraw-nova-docker/blob/experiment... - Result: https://github.com/anteloc/ldraw-nova-docker/blob/experiment...