Most Intelligent AI Model in 2026? What the Rankings Actually Say

As of 14 October 2026 there is no single answer, and the best-known rankings disagree. On the Artificial Analysis Intelligence Index, version 4.3.2, Claude Opus 5.5 leads at about 58 out of 100, ahead of Claude Sonnet 5.5 and Meta's Muse Wonder Full at about 56. On Epoch AI's Capabilities Index the top five models are in a statistical tie. On Scale AI's text-only Humanity's Last Exam board, GPT-6 Astra is first. Which model is "most intelligent" depends on the ruler, and the rulers keep changing. This article gives all three, then checks the claim that started the argument: OpenAI's 3 September description of GPT-6 Astra as "the world's most intelligent and aligned model".

Which AI model is the most intelligent right now?

We read each of these on 14 October 2026:

• Artificial Analysis Intelligence Index v4.3.2, out of 100: Claude Opus 5.5 57.6, Claude Sonnet 5.5 56.0, Muse Wonder Full 55.8, Claude Fable 5.1 53.35, GPT-6 Astra 52.67, Gemini 4 Argon 52.56
• Epoch AI Capabilities Index: Claude Opus 5.5 167.3, GPT-6 Astra 166.5, GPT-6.1 Sol 166.1, Claude Sonnet 5.5 165.0, Claude Fable 5.1 164.7, with overlapping 95% intervals
• Scale AI Humanity's Last Exam, text only: GPT-6 Astra 54.2%, Gemini 3.1 Pro 47.3%, Claude Fable 5.1 46.8%, Claude Sonnet 5.5 46.6%
• Artificial Analysis Humanity's Last Exam, text only: Claude Opus 5.5 61.4%, Claude Fable 5.1 59.1%, GPT-6 Astra 54.7%
• ARC-AGI-3 (ARC Prize Foundation): GPT-6 Astra 62.7% on the neutral harness and 99.9% on OpenAI's own harness

The lists disagree because the rulers differ. Artificial Analysis grades its Humanity's Last Exam run with one automatic judge and Scale AI uses another, which is part of how Claude Fable 5.1 scores 59.1% on one board and 46.8% on the other. Epoch AI publishes an interval for every model, and the intervals of its top five overlap, so the order among them is not settled. And the Intelligence Index itself was rebuilt five times between 6 August and September: Fable 5.1 scored 66 on it on 1 September and 53 under version 4.3 without changing at all. Our explainer on what the Artificial Analysis Intelligence Index actually measures sets out what is inside it and how it has changed.

How a "most intelligent" claim fared on its own table

On 3 September 2026 OpenAI announced GPT-6 Astra and described it, in the first line of the announcement, as "the world's most intelligent and aligned model". The claim travelled everywhere in the following forty-eight hours. It is worth pausing on, not because it is obviously wrong, but because the evidence that would settle it is printed further down the same page — and it does not say what the headline says.

What the announcement's own table shows

The announcement carries a long comparison table setting Astra against GPT-5.6 Sol, Claude Fable 5.1, Claude Fable 5, Claude Opus 5 and Gemini 3.8 Flash across roughly thirty evaluations. Most of those are single-skill tests: a coding benchmark, a science question set, a cybersecurity exercise. Exactly one row is a composite measure of general capability — the Artificial Analysis Intelligence Index, an independent aggregate that blends reasoning, coding, mathematics and knowledge evaluations into one figure.

That row reads as follows.

• Claude Fable 5.1 — 65.7
• Claude Opus 5 — 63.1
• Claude Fable 5 — 62.1
• GPT-6 Astra — 61.2
• GPT-5.6 Sol — 60.9
• Gemini 3.8 Flash — 58.7

Those are the values as the table printed them and as we recorded them when this article was first published, and they come from an earlier version of the index. Artificial Analysis's own 1 September article gives matching whole numbers for four of the models (66, 63, 62 and 61). Under version 4.3.2, which it publishes today, the same models read 53.35 (Claude Fable 5.1), 50.78 (Claude Opus 5), 49.63 (Claude Fable 5) and 46.97 (GPT-5.6 Sol). The index was rebuilt after the table was printed; the order of the three Claude models shown here is the order that mattered on 3 September, not now.

Fourth of six, on the only line in the table that attempts to summarise general capability in a single number, published by the company making the claim. Artificial Analysis, which maintains the index independently, separately called Claude Fable 5.1 the highest score it had measured when it published its own results on 1 September.

A second row points the same way. On Humanity's Last Exam — a set of expert-written questions across dozens of fields, run with tools — the table gives Astra 57.2 per cent against 65.0 for Fable 5.1, 63.8 for Fable 5 and 63.6 for Opus 5. Astra is last of the four models reported. This one is worth more than the others because Anthropic publishes identical figures for those three Claude models on the same evaluation, so the comparison is not contaminated by the two labs running the test differently. The exam itself, and why its top score changes with the board and the tool setting, is explained in our guide to the Humanity's Last Exam leaderboards.

Why the claim is still not a lie

It would be easy, and wrong, to file this as marketing caught out. Astra leads the comparison table on most of its individual rows, and several of those leads are large. It reports 96.0 per cent on GPQA Diamond, a set of graduate-level science questions; 97.6 per cent on FrontierMath Tier 4; 57.9 per cent on Terminal-Bench 4.0 against 55.8 for Fable 5.1; 100 per cent on ExploitBench, up from 78.5 for its predecessor; and 100 per cent on long-context retrieval at 256K to 512K tokens. It also helped establish two new results on the spacing of prime numbers, improving a bound that had stood for more than eighty years.

So there is a defensible reading of "most intelligent" on which the claim holds, and a defensible reading on which it does not, and nothing in the announcement says which reading is meant. That is the actual finding here. The sentence cannot be checked, because the scale is missing.

Why a composite index is the closest thing to a fair question

The reason the Intelligence Index row deserves more weight than the thirty rows around it is the same reason a Full-Scale IQ exists at all. Any single test measures a narrow thing. A composite built from several different tests measures something broader, and if the underlying abilities are related, the composite carries real information that no single subtest does. Epoch AI's Capabilities Index is another composite of this kind; what it is built from and why it is not an IQ are set out in the Epoch Capabilities Index explained.

In people, they are related. Performance across vocabulary, spatial reasoning, working memory and arithmetic correlates positively — the pattern known as the general factor, which we set out in our explainer on what the g factor actually is. That correlation is what licenses a single summary number, and it is why an IQ generalises beyond the specific items on the paper.

Machine abilities do not hang together like that. Results depend on conditions as much as on any underlying quantity: on the coloured-grid puzzles of ARC-AGI-3, which untrained adults solve reliably, GPT-6 Astra scored 62.7 per cent on the ARC Prize Foundation's neutral harness and 99.9 per cent on a harness supplied by OpenAI, as our guide to the two ARC-AGI-3 scores for one model explains. Because the abilities are not correlated, any composite is a weighting decision — a statement about which capabilities the index author thinks matter — rather than a measurement of an underlying quantity. Change the weights and the ranking changes. A different kind of ranking, a rescaled count of right answers on a 35-puzzle test, is dissected in our guide to AI scores on the Mensa Norway test.

That is a genuine limitation of the index, and it cuts both ways. It is a reason not to treat 65.7 versus 61.2 as a verdict on which system is cleverer. It is also the reason "most intelligent" cannot be rescued by pointing at individual wins: if there is no single quantity being measured, there is no single ranking to top. GPQA Diamond, one of the best-known science benchmarks behind such rankings, was dropped from the index in version 4.2; it is explained in our guide to the GPQA Diamond score.

The same rule this section applies to human scores

Everything above is one rule applied to a new subject. A number becomes a measurement when three things are stated: the scale it sits on, the population it is compared against, and the conditions under which it was produced. Strip any of the three and what is left is a figure with a decimal point in it. For a human-judged test of AI that reports a win rate instead of an index, see our guide to the 2026 PNAS Turing test study.

• An IQ of 130 means nothing until you know the standard deviation — it is the 97.7th percentile on a 15-point scale and the 96.9th on a 16-point scale, as we set out in our piece on why one percentile has three numbers.
• A benchmark percentage means nothing until you know the harness, the effort setting and whether the figure is a single run or the best of several.
• A composite means nothing until you know what went into it and how it was weighted.
• And "most intelligent", with none of the three, means nothing at all.

None of this is a reason to dismiss the model or the benchmarks. Frontier evaluation in 2026 is a serious enterprise producing genuinely informative numbers, and the Artificial Analysis index is a reasonable attempt at the hardest part of it. The point is narrower: the superlative on the front of the announcement is doing none of that work, and the reader who wants to know what changed on 3 September has to scroll past it to the table. Another AI measure, the length of task an agent completes, is explained in METR time horizon explained.

The useful summary of the launch is unglamorous and entirely supportable from the published tables. GPT-6 Astra is the strongest system yet released on a long list of specific, valuable capabilities, and on the one general-capability composite in its own announcement it is fourth. Both halves of that sentence are printed by OpenAI. Only one of them travelled. For a measured score that states its scale and its population, take the IQ Metrics test.

Common questions