On October 27, 2026, Artificial Analysis — the independent benchmarking company behind one of the most cited AI rankings — launched version 5 of its Intelligence Index. Most of it now runs on private tests, and the ranking moved. Here is the new list, what changed and what it means when you pick a model.
All language models on AIVerso with a published v5 score. Click a column header to sort.
| # | Model | Provider | v5 | Before (v4) | Change | API price (per 1M) | Open weights | AIVerso score |
|---|---|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 | Anthropic | 65 | — | new | $8/$40 | No | 99 |
| 2 | Gemini 4 Carbon | 60 | — | new | $2/$10 | No | 98 | |
| 3 | Claude Opus 5.5 | Anthropic | 59 | 58 | +1 | $4/$20 | No | 98 |
| 4 | GPT-6 Astra | OpenAI | 57 | 53 | +4 | $10/$50 | No | 98 |
| 5 | GPT-6.1 Sol | OpenAI | 55 | 52 | +3 | $2/$10 | No | 97 |
| 6 | Claude Sonnet 5.5 | Anthropic | 55 | 56 | -1 | $2/$10 | No | 97 |
| 7 | GPT-6.1 Terra | OpenAI | 51 | — | new | $1/$5 | No | 96 |
| 8 | Gemini 4 Argon | 50 | 53 | -3 | $2/$10 | No | 96 | |
| 9 | Kimi K3.1 Pro | Moonshot AI | 46 | 55 | -9 | $2/$10 | Yes | 95 |
| 10 | Muse Wonder 1.0 | Meta | 44 | 52 | -8 | $4/$20 | No | 93 |
| 11 | Muse Ember | Meta | 36 | — | new | $0.10/$0.50 | No | 91 |
| 12 | Claude Haiku 5.5 | Anthropic | 36 | 43 | -7 | $0.10/$0.50 | No | 91 |
| 13 | Grok 4.7 | xAI | 35 | 46 | -11 | $2/$6 | No | 91 |
| 14 | Tencent Hy4 | Tencent | 35 | — | new | $0.90/$2.70 | No | 91 |
| 15 | Qwen 3.8-Max | Alibaba Cloud | 34 | 45 | -11 | $2/$6 | Yes | 90 |
| 16 | DeepSeek V4.1 Flash | DeepSeek | 33 | 45 | -12 | $0.30/$1.20 | Yes | 90 |
| 17 | GLM-5.3 | Zhipu AI | 32 | 45 | -13 | $1.40/$4.40 | Yes | 90 |
| 18 | Inkling 2 Small | Thinking Machines | 31 | — | new | $0.30/$1 | No | 89 |
| 19 | MiMo-V2.6-Pro | Xiaomi | 30 | 46 | -16 | $0.44/$0.87 | Yes | 89 |
| 20 | MiniMax M3.1 Flash | MiniMax | 29 | 43 | -14 | $0.15/$0.60 | Yes | 88 |
| 21 | GPT-6 Luna | OpenAI | 28 | 38 | -10 | $0.10/$0.50 | No | 88 |
| 22 | Mistral Large 4 | Mistral AI | 26 | 38 | -12 | $0.70/$2.10 | Yes | 87 |
| 23 | Muse Spark Open | Meta | 25 | 39 | -14 | $1.25/$4.25 | Yes | 87 |
| 24 | Command A+ | Cohere | 6 | 13 | -7 | — | Yes | 78 |
v5 = Artificial Analysis Intelligence Index v5 (highest-scoring effort setting). Before = last v4 value. API prices are list prices in USD per 1M input/output tokens.
Claude Fable 5.2 (65), Gemini 4 Carbon (60), GPT-6.1 Terra (51), Muse Ember (36), Tencent Hy4 (35), Inkling 2 Small (31) had no v4 score yet.
9 evaluations in four areas, each worth 25% of the index. 60% of the weight comes from tests with private questions or held-back answers. Removed from the index: GDPval-AA v2.1, SciCode and AA-LCR v1.1.
| Area | Evaluation | Weight | Status |
|---|---|---|---|
| Knowledge work | AA-Briefcase | 15% | Kept from v4 |
| Knowledge work | AutomationBench-AA | 5% | Kept from v4 |
| Knowledge work | GDP.pdf | 5% | Kept from v4 |
| Coding | FrontierCode 2.0 | 15% | New |
| Coding | Terminal-Bench 4.0 | 10% | Kept from v4 |
| Science | Terminal-Bench-Science | 15% | New |
| Science | CritPt v2 (hard subset) | 10% | Upgraded |
| General | AA-Omniscience v1.1 | 15% | Upgraded |
| General | HLE-Diamond | 10% | Upgraded |
| Evaluation | Leader | Score | Next best |
|---|---|---|---|
| FrontierCode 2.0 | Claude Opus 5.5 | 46% | GPT-6 Astra, GPT-6.1 Terra and Claude Sonnet 5.5: 42% |
| Terminal-Bench-Science | GPT-6 Astra | 63% | Claude Opus 5.5: 59% · GPT-6.1 Sol: 58% |
| HLE-Diamond | Claude Opus 5.5 | 68% | GPT-6 Astra: 64% |
| CritPt v2 (hard subset) | GPT-6 Astra | 72% | GPT-6.1 Sol: 69% · Claude Opus 5.5: 61% |
| Terminal-Bench 4.0 | Kimi K3.1 Pro | 67% | — |
| AutomationBench-AA | Gemini 4 Argon | 78% | — |
| GDP.pdf | Muse Wonder Full | 33% | all-pass rate; Muse Wonder Full is only available in Muse Code plans |
| AA-Briefcase | Claude Sonnet 5.5 | 1812 Elo | — |
| AA-Omniscience v1.1 | Claude Opus 5.5 | 48 | — |
For language models, the AIVerso score follows the Intelligence Index: since v5 we use score ≈ 78 + 0.36 × index (the top model gets 99). Image, video and voice tools are rated with the Artificial Analysis arenas instead. Prices, free plans and features are shown separately on every model page, so you can weigh them yourself.
A composite score from Artificial Analysis, an independent company that benchmarks AI models. Version 5 combines 9 evaluations in four equally weighted areas — knowledge work, coding, science and general reasoning & knowledge — into one number per model.
Version 5 was published on October 27, 2026. It adds FrontierCode 2.0 and Terminal-Bench-Science, upgrades CritPt v2 (hard subset), AA-Omniscience v1.1 and HLE-Diamond and drops GDPval-AA v2.1, SciCode and AA-LCR v1.1. For the first time, most of the index (60%) is based on tests with private questions or held-back answers.
The top model you can use today is Claude Fable 5.2 (65), followed by Gemini 4 Carbon (60) and Claude Opus 5.5 (59).
Most of v5 runs on new, harder and private tests, and open-weight models scored clearly lower on them (minus 7–16 points on AIVerso’s list). According to Artificial Analysis, the gap between the leader and the best open-weights model widened from 3 to 12 points.
No. The AIVerso score (0–100) also covers image, video, voice and app tools. For language models we calibrate it against the Artificial Analysis Intelligence Index, so the order of the top text models follows the index; price, free access and features are shown separately.
Source: Artificial Analysis — Announcing Intelligence Index v5 and the model leaderboard. AIVerso is not affiliated with Artificial Analysis. Last updated: November 11, 2026.