**[Local Models Are Getting Wild: Qwen 3.8-27B Matches DeepSeek V4 Flash in Benchmark Scores]** According to monitoring by Dongcha Beating, Artificial Analysis has finally released the independent benchmark score for Qwen 3.8-27B: Intelligence Index 52 points. This dense model with only 27B parameters directly matches the scores of DeepSeek V4 Flash 0731 and GPT-5.6 Luna Max, trailing just 1 point behind DeepSeek V4 Pro 0813 and GLM-5.2, which scored 53 points. The previous generation Qwen 3.6-27B only scored 38 points. What’s even more astonishing is its Agent capabilities. Qwen 3.8-27B achieved an Agentic Index of 51 points, surpassing DeepSeek V4 Pro’s 50 points, DeepSeek V4 Flash’s 48 points, and GPT-5.6 Luna’s 47 points. This metric specifically tests the model’s ability to utilize tools, plan steps, and complete complex tasks in sequence. However, the 52-point score comes at a cost. During the full Intelligence Index test, Qwen 3.8-27B generated approximately 160 million output tokens in total, clearly falling into the category of "intelligence through token stacking." The default inference mode for Qwen is xhigh, with support for medium, low, and no-thinking modes. Currently, Artificial Analysis has not specified which inference mode was used in the test, so it’s unclear if this score reflects the default xhigh setting. [Original Link]