Qwen 3.8 27B Matches GPT-5.6 Luna on Intelligence Index

👁 117 reads

Alibaba’s Qwen 3.8 27B scored 52 on the Artificial Analysis Intelligence Index as of 17 August 2026, matching GPT-5.6 Luna at maximum settings and finishing just one point behind both GLM-5.2 and DeepSeek V4 Pro 0813.

The size contrast is striking. GLM-5.2 carries 753 billion parameters and DeepSeek V4 Pro 0813 runs at 1.7 trillion parameters, while Qwen 3.8 27B weighs in at 27 billion. The size of Luna is not publicly disclosed, though it is widely assumed to be substantially larger than 27B. All three rival models were running at their maximum settings when they posted their scores.

Developer Simon Willison, writing on his blog, described Qwen 3.8 27B as “a truly astonishing model.” Willison had separately noted that the model “defaults to wildly overthinking things,” suggesting its benchmark efficiency does not come without trade-offs in everyday use.

// read moreQwen 3.8 vs GPT-5.6 Luna: 7x Price Gap, Same ContextBenchmarksQwen 3.8 27B Hobbled by Its Default Overthinking SettingBenchmarksQwen 27B: Style Training Changed AI Behavior UnexpectedlyBenchmarks

Kinetyra