Qwen 3.6 27B on DeepSWE
The Qwen 3.6 27B model achieved a score of 1.79% on the DeepSWE benchmark, placing it in 18th out of 20 models. This benchmark run, which took 70 hours to complete, utilized an RTX6000 Pro Blackwell GPU and a 262k context window. Despite a community reputation for verbosity, the model's output tokens were comparable to similar models, and it is considered a strong local option compared to leading closed-source models like Kimi. AI
IMPACT Provides a performance benchmark for an open-source model, indicating its capabilities relative to other models in the local LLM ecosystem.