Glonce

Qwen 3.8 27B scores 52 on the Artificial Analysis Intelligence Index

modelsThe Glonce editorsSimon Willison🔥76Glonce rating76Impact68Novelty62Relevance98Surprise72Players78Credibility82

Qwen 3.8 27B scored 52 on the Artificial Analysis Intelligence Index. That result ties GPT-5.6 Luna (max), a model whose parameter count is not disclosed but is presumed to be far larger than 27 billion. Qwen 3.8 27B also finished just one point behind two other models measured on the same index: GLM-5.2 (max), listed at 753 billion parameters and far larger than Qwen, and DeepSeek V4 Pro 0813 (max), listed at 1.6 billion parameters and smaller than Qwen.

Simon Willison, writing in a link blog post on 17 August 2026, called Qwen 3.8 27B "a truly astonishing model" given how close its score sits to other models, some built at a much larger scale and one much smaller. He did not name the lab behind Qwen 3.8 27B or the other models in the comparison, explain how the Intelligence Index is calculated, or say when the underlying scores were measured; the post presents the index numbers as a straight comparison across models. In a separate post the day before, Willison had noted that Qwen 3.8 27B tends to overthink problems even as it scores well.

Key facts

  • Qwen 3.8 27B scored 52 on the Artificial Analysis Intelligence Index, matching GPT-5.6 Luna (max).
  • It finished just one point behind GLM-5.2 (max) and DeepSeek V4 Pro 0813 (max) on the same index.
  • GLM-5.2 (max) is listed at 753 billion parameters and DeepSeek V4 Pro 0813 (max) at 1.6 billion parameters, against Qwen 3.8 27B's 27 billion.
  • Luna's parameter count is undisclosed but presumed much larger than Qwen 3.8 27B's 27 billion.
  • Simon Willison called Qwen 3.8 27B "a truly astonishing model" for closing that gap against far bigger systems.

Why it matters

A 27 billion parameter model landing within one point of models of very different sizes on the same benchmark is notable because it runs against the usual assumption that a higher Intelligence Index score requires a much bigger model. Qwen 3.8 27B ties or nearly matches GPT-5.6 Luna (max), GLM-5.2 (max) at 753 billion parameters and far larger than Qwen, and DeepSeek V4 Pro 0813 (max) at 1.6 billion parameters and far smaller than Qwen, as reported in the source, all on the same index.

Who it affects

Anyone choosing between AI models based on published benchmark scores, including developers weighing a smaller model against much larger competitors, and researchers tracking how model size relates to Intelligence Index performance.

How to use it

The source gives no pricing, licensing, or deployment details for Qwen 3.8 27B or the models it is compared against. It reports only the Intelligence Index scores and the parameter counts named above; no other usage information is provided.

How solid is it

The post is a short link blog entry pointing to the Artificial Analysis Intelligence Index results, without explaining how the index is computed, without naming the lab behind any of the four models, and without a date for when the underlying scores were measured. It rests on a single named source, Simon Willison, citing that index.

Risks and caveats

The comparison covers one benchmark index only, and the source supplies no methodology for it. The 1.6 billion parameter figure given for DeepSeek V4 Pro 0813 (max) is reported exactly as the source states it, with no further detail to check it against. No developer or lab is named for any of the four models, so the claims cannot be traced to an original announcement from that source alone.

“Qwen 3.8 27B is a truly astonishing model.”

— Simon Willison