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

Bulletin. Artificial Analysis has posted a new entry to its Intelligence Index. Qwen 3.8 27B, a comparatively small open model, has recorded a score of 52 on the benchmark. That figure matches the score posted by GPT-5.6 Luna (max) and trails by a single point the scores of GLM-5.2 (max) and DeepSeek V4 Pro 0813 (max).

The comparison is notable chiefly for scale. GLM-5.2 carries 753 billion parameters. DeepSeek V4 Pro 0813 carries 1.7 trillion. Qwen 3.8 27B, by contrast, runs at 27 billion parameters, a small fraction of either. Luna’s parameter count has not been disclosed, though commentary accompanying the result presumes it falls into a similarly large, proprietary class.

Simon Willison, reporting the figures, frames the result as evidence that a much smaller open-weight model can now register benchmark scores on par with systems many times its size, some of them closed and commercially maintained. The index itself, maintained by Artificial Analysis, aggregates model performance into a single comparative number, and this entry places Qwen 3.8 27B within one point of the current upper cluster.

No further detail on training data, method, or cost has been included in the cited material. The observatory notes the figures as reported and awaits fuller documentation before any wider comparison of efficiency or capability can be drawn.