Artificial Analysis Index v5: Opus 5.5 leads at 58, open models trail

Artificial Analysis released Intelligence Index v5 on October 27, a rebuilt version of its widely cited model ranking. Claude Opus 5.5 at max effort leads with a score of 58, ahead of GPT-6 Astra at 57 and GPT-6.1 Sol at 55. The bigger change is the gap behind them: Artificial Analysis says the best open-weight model now trails the leader by 12 points, up from 3 in the last version.

This matters because many teams pick models from this one number. A new version with different tests can reorder the list, and this one is built so that models are harder to tune for.

What changed in v5

Artificial Analysis calls v5 its biggest upgrade since v4 launched in January. The index now has nine evaluations in four categories, each worth 25%: knowledge work, coding, science and general knowledge. Two evaluations are new, FrontierCode 2.0 and Terminal-Bench-Science. Three were removed: GDPval-AA, SciCode and AA-LCR.

About 60% of the weight now sits on tests with private datasets or held-back answers. Before, most of the index used public tests. Private tests are harder to train on, which is the point. The methodology page marks five of the nine evaluations as private.

Artificial Analysis also launched a separate Progress Index v1, which tracks models released since November 2022 on one continuous scale. It says the Intelligence Index stays the main measure of current models.

Who leads, and on what

Anthropic and OpenAI hold the top five places, according to Artificial Analysis. Gemini 4 Argon at high effort scores 50. The leaders are split by test:

• Opus 5.5 leads FrontierCode 2.0 (46%), HLE-Diamond (68%) and AA-Omniscience v1.1.
• GPT-6 Astra leads Terminal-Bench-Science (63%) and CritPt v2 (72%).
• Claude Sonnet 5.5 leads AA-Briefcase, Kimi K3.1 Pro leads Terminal-Bench 4.0 (67%), Gemini 4 Argon leads AutomationBench-AA (78%) and Muse Wonder Full leads GDP.pdf.

Artificial Analysis notes that no single model leads them all. For a head-to-head on price and limits, see our comparison of Gemini 4 Argon and Opus 5.5.

Open-weight models fall further behind

The top open-weight model is Kimi K3.1 Pro at 46, but Artificial Analysis says its weights are not yet released. Among models whose weights are out, DeepSeek V4.1 Flash 1017 and GLM-5.3 lead at 32, which is 26 points behind the leader. Claude Haiku 5.5, a closed model, scores 36.

The gap is widest on HLE-Diamond, where Kimi scores 37% against 68% for Opus 5.5, and on Terminal-Bench-Science, 34% against 63%. Our earlier piece on Kimi K3.1 Pro pricing has what the model costs through its API.

Cost per task and effort settings

Artificial Analysis reports cost per index task for the ten highest-scoring models. GPT-6.1 Terra at max effort is cheapest at $0.96. Claude Fable 5.1 at max is the most expensive at $13.04. Opus 5.5 costs $10.04 per task, GPT-6 Astra $6.24 and GPT-6.1 Sol $1.27. By those numbers Sol scores within four points of Opus for about an eighth of the cost.

Effort settings matter more than before. Sonnet 5.5 spans 29 points from low to max effort, up from 20 in v4.3. Opus 5.5 scores 35 at low effort and 58 at max. Median output per task rose from 35,000 to 59,000 tokens, because agentic tests that run for many steps make up 55% of the weight.

What to keep in mind

These are scores from the benchmark's own publisher, on a test it wrote and mostly keeps private, so outsiders cannot check the questions. A single index also hides the split by task: if you build a coding agent, the FrontierCode and Terminal-Bench columns say more than the headline number.

Artificial Analysis publishes the full model table on the article page. Scores move as providers update models, so check it before you pick one.