Artificial Analysis rebuilt its Intelligence Index around private test data, and the open-weights gap it measures widened from 3 points to 12
AspireAI community · Artificial Analysis
*The most-watched open model leaderboard just changed how it measures, and the change moved the numbers.* Artificial Analysis published Intelligence Index v5 on October 27, the biggest rework of the composite since v4 launched in January. Sixty percent of the index now rests on evaluations with private test sets or held-out answers. Two new benchmarks enter: Cognition's FrontierCode 2.0 (132 tasks in real open-source repositories, graded on whether a maintainer would merge the agent's change) and Terminal-Bench-Science (70 agentic scientific research tasks run in a terminal). Claude Opus 5.5 at max effort leads the index at 58, ahead of GPT-6 Astra at 57 and GPT-6.1 Sol at 55.
The open-weights gap is the number worth remembering: Kimi K3.1 Pro is the top open-weights model at 46, but its weights are still unreleased, and among models with released weights DeepSeek V4.1 Flash 1017 and GLM-5.3 lead at 32 - below Claude Haiku 5.5 at 36. The honest caveat is that no single model leads all nine evaluations; Claude Opus 5.5 leads three and five other models lead the rest. Cost per task varies by more than 13x across the top ten, and the agentic evaluations now make up 55 percent of the index weight while consuming a median of about 90 percent of output tokens, with median output per task up from 35k to 59k tokens.
Read the story and explore more from the AspireAI community.
Two sparse-autoencoder analyses have resisted scaling: matching semantically similar features across layers, and compressing large feature circuits into interpretable supernodes. This paper argues both are the same underlying problem, estimating semantic distances between SAE features that sit on different activation manifolds. Its representation treats each feature not as a single decoder vector but as an activation-weighted distribution over the hidden states expressing it, projects those distributions into a shared reference space, and compares them with Wasserstein distance. The representation is proved invariant to activation rescaling and stable under perturbation, and recovers true matches under finite-sample margin conditions. Empirically it outperformed decoder-vector and LLM-based baselines and captured subtle functional distinctions between related features, and it compressed large feature circuits into interpretable supernodes automatically. The limitation is that this is interpretability tooling: the paper demonstrates better feature matching and circuit compression, not that downstream interpretability claims become more accurate.