On this page
On October 28, 2026 at 10:11 Asia/Shanghai, @ArtificialAnlys announced that the Artificial Analysis Intelligence Index has moved to v5. For the first time, a majority—about 60%—rests on private test sets and held-out answers. Related reading (Coding Agent Index v2.0 / FrontierCode v1.1—not Intelligence Index v5 / FrontierCode 2.0): Coding Agent Index Moves to v2.0 With FrontierCode — Claude Code + Opus 5.5 Tops the New Board.
These benchmarks have not been re-run on this site; the scores below are Artificial Analysis’s published figures.
What changed in v5
Intelligence Index v5.0 lists nine evaluations: AA-Briefcase v1.1, AutomationBench-AA, Terminal-Bench 4.0, FrontierCode 2.0, AA-Omniscience v1.1, GDP.pdf, HLE-Diamond, CritPt v2 (Hard Subset), and Terminal-Bench-Science 0.1.
New: FrontierCode 2.0 and Terminal-Bench-Science. Upgraded: HLE-Diamond (replacing Humanity's Last Exam), AA-Omniscience v1.1, and CritPt v2 hard subset. Kept: AA-Briefcase, AutomationBench-AA, GDP.pdf, and Terminal-Bench 4.0. Removed: GDPval-AA v2.1, SciCode, and AA-LCR v1.1.
How the score is weighted
Four 25% quadrants:
- Knowledge Work 25%: AA-Briefcase 15%, AutomationBench-AA 5%, GDP.pdf 5%
- Coding 25%: FrontierCode 2.0 15%, Terminal-Bench 4.0 10%
- Science 25%: Terminal-Bench-Science 15%, CritPt v2 hard subset 10%
- General 25%: AA-Omniscience v1.1 15% (accuracy 10% + non-hallucination 5%), HLE-Diamond 10%
The coding slice here is Intelligence Index’s FrontierCode 2.0, not the older Coding Agent Index article linked above.
Leaderboard highlights (max)
Claude Opus 5.5 (max) leads at 58, then GPT-6 Astra (max) 57 and GPT-6.1 Sol (max) 55. Anthropic and OpenAI take the top five; Gemini 4 Argon (high) sits at 50.
Kimi K3.1 Pro scores 46—best open-side figure here (weights not released), trailing the leader by 12 (the gap was 3 on v4.3). Among released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) both sit at 32. Claude Haiku 5.5 (max) at 36 still clears those two.
FrontierCode 2.0
FrontierCode 2.0 is a Cognition private board in the Coding quadrant: 132 real open-source repo tasks, scored by whether a maintainer would merge. Blocking failures or looking up answers score 0; tasks stay private. Artificial Analysis runs each task three times on Stirrup.
Opus 5.5 (max) leads at about 46% (the evaluation page may show 46.4%). Astra, GPT-6.1 Terra, and Sonnet 5.5 (max) cluster around 42% (site figures near 42.4%). Best open-weight listed: DeepSeek V4.1 Flash 1017 (max) at 28%.
For agent pricing and plan limits, see Claude Code or Codex? How Their Prices and Usage Limits Compare.
Cost per Index task and token load
Per Intelligence Index task: about $10.04 for Opus 5.5 (max), $6.24 for Astra (max), and $1.27 for GPT-6.1 Sol (max)—Sol stays within about four points of Opus at roughly one-eighth the cost. Other top-ten costs: Terra (max) about $0.96; Fable 5.1 (max) about $13.04 (more than 13× Sol).
Versus v4.3, median output tokens per task rose from about 35k to 59k (roughly +2/3). Effort tiers matter more: Opus low at 35 versus max at 58.
Another Artificial Analysis eval pairs capability with spend: Code2Video-AA: How Models Design Videos in Code — and What It Costs. Provider choice can move speed and bill on the same open weights—see Same Open Model, Different API: Why Speed and Cost Both Change.
Progress Index v1 is a separate product
Progress Index v1 is an IRT long scale comparable back to November 2022. It tracks progress, not a drop-in for today’s frontier table. For current use and current frontier rankings, Artificial Analysis still points to the Intelligence Index. Composite scores across Intelligence Index versions (v4.3 vs v5) are not directly comparable.
Figures follow Artificial Analysis’s posts and site. For reference only.