Artificial Analysis Intelligence Index
Most fields on a gallery card describe the model itself, such as its attention type, layer mix, or KV-cache size. I added the Artificial Analysis Intelligence Index to provide a separate performance reference. It answers a different question. How did the released model score on a shared set of evaluations?
The card shows this information in two lines. Total score is copied from the matching Artificial Analysis model page. Profile is my compact diagnostic view across Knowledge Work, Coding, Science, and General, using the v5.0 category weights.
There is an important distinction between them. The total is the official Intelligence Index. The profile is specific to this gallery and should not be used to reconstruct the total.
Gallery method
Intelligence Index v5.0
Category weights
Knowledge Work, Coding, Science, and General each contribute 25%
Gallery data snapshot
2026-11-02
What the two fields mean
Total score is the overall Intelligence Index shown on the matching Artificial Analysis model page. If that page reports an estimate while an independent evaluation is pending, the gallery records the estimate.
The November 1, 2026 refresh added v5.0 scores throughout the gallery. A model without a published v5.0 total is recorded as N/A, even if it previously had a v4.3.2 score. Each model shows a separate index card for every version with an available total or profile.
The gallery data also retains the earlier v4.3.2 totals and profiles in separate fields. Those values preserve the October 24, 2026 published snapshot. When both versions are available, the model shows a v5.0 index card followed by a v4.3.2 index card. Selecting AA Index v4.3.2 score or AA Index v5.0 score changes the ranking while keeping both cards visible. Missing scores appear last in either ranking. The comparison tool follows the selected score version and uses v5.0 for other sort modes.
This snapshot covers all 110 gallery entries. Of these, 87 have matching AA pages, 15 have v5.0 totals, and 17 have at least one available profile category. The other 23 entries have no matching AA page.
Profile is calculated from the benchmark values exposed on the model page. Each category is shown on a 0–100 scale. Missing inputs make that category N/A; a reported zero remains zero.
For example, Mistral Large 4 has a total of 26.0 alongside Knowledge Work 40.6, Coding 18.9, Science 13.0, and General 31.7 in this snapshot.
AA freezes the Briefcase Elo used in its official total. The page’s published Elo can subsequently change, so the average of the gallery’s four profile values can differ from the official total. Rounding also contributes to small differences. I keep the profile for comparing strengths across categories and always copy the total directly from AA.
What goes into v5.0
The nine evaluations are grouped as follows. Weights here refer to the full index.
- Knowledge Work. AA-Briefcase v1.1 contributes 15%, AutomationBench-AA 5%, and GDP.pdf 5%.
- Coding. FrontierCode 2.0 contributes 15%, and Terminal-Bench 4.0 contributes 10%.
- Science. Terminal-Bench-Science 0.1 contributes 15%, and the hard subset of CritPt v2 contributes 10%.
- General. AA-Omniscience v1.1 contributes 10% for accuracy and 5% for non-hallucination rate. HLE-Diamond contributes 10%.
To calculate a profile category, I divide these weights by its 25% share. For example, Coding is 60% FrontierCode and 40% Terminal-Bench. Briefcase Elo is normalized as (Elo - 500) / 2000, clipped to 0–1, before weighting. The resulting category values are multiplied by 100.
Why the snapshot date matters
Artificial Analysis revises both benchmark composition and model results. For that reason, the gallery stores the date of its data pull. The current snapshot on this page is 2026-11-02.
The date matters in model comparisons. Scores collected under different index versions may reflect changes in the evaluation suite as well as changes in the models. The refresh checks the exact model slug and index version before updating scores. A fetch or parsing error stops the write so that older results cannot silently remain in a new snapshot.
What the score can tell us
I use the index as a broad check on a model release. A score difference cannot identify which design choice caused it. Training data, post-training, tool use, reasoning settings, and test-time compute can all influence the result alongside architecture.
The v5.0 suite is primarily based on English text tasks. Image-capable models also receive images in some evaluations, including HLE-Diamond and GDP.pdf. It does not provide a complete assessment of multimodal or multilingual capability.
For architecture comparisons, I therefore use the index as supporting context. The architecture fields and the original technical report remain the better sources for understanding how a model is built.