Can I compare Intelligence Index v5 scores with v4 scores?

Artificial Analysis treats Intelligence Index v5 as a major revision with a different benchmark set, so v5 numbers form a new series. The announcement compares behaviour against v4.3 — for example, the open-weights gap grew from 3 to 12 points — but it does not say v5 and v4.3 scores are directly comparable. Compare models within the same version.

Which open-weights model scores highest on Intelligence Index v5?

Kimi K3.1 Pro scores 46 on Intelligence Index v5, the top open-weights result, but Artificial Analysis notes its weights are not yet released. Among models whose weights are already out, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) lead at 32, which is 26 points behind Claude Opus 5.5 (max) at 58.

Which model is cheapest to run on Intelligence Index v5?

Among the top ten models, GPT-6.1 Terra (max) is the cheapest at $0.96 per Intelligence Index v5 task and Claude Fable 5.1 (max) the most expensive at $13.04. GPT-6.1 Sol (max) costs $1.27 per task and scores 55, within 4 points of Claude Opus 5.5 (max), which costs $10.04 per task.

Does one model win every benchmark in Intelligence Index v5?

No model leads all nine Intelligence Index v5 evaluations. Claude Opus 5.5 tops FrontierCode 2.0 (46%) and HLE-Diamond (68%), GPT-6 Astra leads Terminal-Bench-Science (63%) and CritPt v2 (72%), Kimi K3.1 Pro leads Terminal-Bench 4.0 (67%), and Gemini 4 Argon leads AutomationBench-AA (78%), according to Artificial Analysis.

Artificial Analysis · 2026-10-27 · major

Artificial Analysis Index v5 — private test sets now carry 60%

Artificial Analysis Intelligence Index v5 adds FrontierCode 2.0 and Terminal-Bench-Science and swaps Humanity's Last Exam for HLE-Diamond. 60% of the Index now uses private data. Claude Opus 5.5 (max) leads at 58.

Artificial Analysis Intelligence Index v5 announcement graphic

Artificial Analysis rebuilds its model leaderboard so most of the score comes from tests models cannot have seen.

Key specs

Private share	60%
Top score	58 (Claude Opus 5.5 max)

Quick facts

Version	v5 (previous: v4.3)
Published	October 27, 2026
Evaluations	9, in four 25% categories
New	FrontierCode 2.0, Terminal-Bench-Science
Upgraded	HLE-Diamond, AA-Omniscience v1.1, CritPt v2 hard subset
Removed	GDPval-AA v2.1, SciCode, AA-LCR v1.1

Benchmarks

Artificial Analysis Intelligence Index v5

Claude Opus 5.5 (max)		58 pts
GPT-6 Astra (max)		57 pts
GPT-6.1 Sol (max)		55 pts
Gemini 4 Argon (high)		50 pts
Kimi K3.1 Pro		46 pts
DeepSeek V4.1 Flash 1017 (max)		32 pts
GLM-5.3 (max)		32 pts

source ↗

What is it?

Intelligence Index v5 is the biggest revision of Artificial Analysis's composite model score since v4 in January. It adds Cognition's FrontierCode 2.0 and the new Terminal-Bench-Science, replaces Humanity's Last Exam with HLE-Diamond, and drops GDPval-AA, SciCode and AA-LCR. A Progress Index v1 launched alongside it to place every language model since November 2022 on one scale.

How does it work?

Nine evaluations are grouped into four equal 25% categories: Knowledge Work, Coding, Science and General. On FrontierCode 2.0, open-source maintainers judge whether an agent's change to a real repository would be merged; Terminal-Bench-Science has 70 agentic tasks across the sciences. Agentic tests carry 55% of the weight but use about 90% of output tokens, and the median model now spends 59k output tokens per task, up from 35k in v4.3.

Why does it matter?

Private and held-out data make the score harder to game by training on public test sets, so the ranking says more about real ability. The rebuild also widens the gap between open and closed models: the best open-weights entry, Kimi K3.1 Pro, trails the leader by 12 points, up from 3 in v4.3. Cost differs a lot too, with GPT-6.1 Sol at $1.27 per task within 4 points of Claude Opus 5.5 at $10.04.

Who is it for?

teams choosing a model, ML researchers tracking evals

Frequently asked questions

Can I compare Intelligence Index v5 scores with v4 scores?: Artificial Analysis treats Intelligence Index v5 as a major revision with a different benchmark set, so v5 numbers form a new series. The announcement compares behaviour against v4.3 — for example, the open-weights gap grew from 3 to 12 points — but it does not say v5 and v4.3 scores are directly comparable. Compare models within the same version.
Which open-weights model scores highest on Intelligence Index v5?: Kimi K3.1 Pro scores 46 on Intelligence Index v5, the top open-weights result, but Artificial Analysis notes its weights are not yet released. Among models whose weights are already out, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) lead at 32, which is 26 points behind Claude Opus 5.5 (max) at 58.
Which model is cheapest to run on Intelligence Index v5?: Among the top ten models, GPT-6.1 Terra (max) is the cheapest at $0.96 per Intelligence Index v5 task and Claude Fable 5.1 (max) the most expensive at $13.04. GPT-6.1 Sol (max) costs $1.27 per task and scores 55, within 4 points of Claude Opus 5.5 (max), which costs $10.04 per task.
Does one model win every benchmark in Intelligence Index v5?: No model leads all nine Intelligence Index v5 evaluations. Claude Opus 5.5 tops FrontierCode 2.0 (46%) and HLE-Diamond (68%), GPT-6 Astra leads Terminal-Bench-Science (63%) and CritPt v2 (72%), Kimi K3.1 Pro leads Terminal-Bench 4.0 (67%), and Gemini 4 Argon leads AutomationBench-AA (78%), according to Artificial Analysis.

Try it

https://artificialanalysis.ai/methodology/intelligence-benchmarking