Kimi K3.1 Pro scores 55 on Artificial Analysis and reaches the cost-performance Pareto frontier

Kimi K3.1 Pro has posted a score of 55 on the Artificial Analysis Intelligence Index and has been placed on the tracker's Pareto line, according to a Hacker News front-page item published on October 23, 2026. The item links directly to Artificial Analysis' analysis page for the model, which charts the release against the current field on intelligence, speed and price. The page tracks the Pro variant of the model.

What the index measures

The Artificial Analysis Intelligence Index, currently at version 4.3.2, aggregates ten evaluations that the site runs independently: AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, Humanity's Last Exam, GDP.pdf, CritPt, AA-Omniscience and AA-LCR v1.1. According to the site, that mix covers agentic knowledge work, research workflows driven inside a terminal, software engineering agents working in open-source codebases, reasoning over professional documents, and long-context medical reasoning, alongside AA-Omniscience, which grades knowledge reliability on a scale from -100 to 100 by giving credit for correct answers, subtracting for hallucinations and treating refusals neutrally.

The index also feeds a cost model. Artificial Analysis derives a per-task cost figure from input, cache-hit, cache-write, reasoning and output token prices, divided by the number of tasks in each evaluation and weighted by that evaluation's share of the index. That cost figure is what makes the frontier comparison possible in the first place.

What the Pareto placement means

The tracker's headline chart plots the Intelligence Index against cost per task. A model sits on the Pareto line when no other tracked model is simultaneously more capable and cheaper per task; anything off the line is dominated by at least one model that beats it on one axis without losing on the other. A placement on the line at a score of 55 therefore means K3.1's capability-per-dollar combination was not bettered by anything else in the tracker's field at the time of the snapshot.

One note on provenance: the 55 figure comes from the title of the Hacker News item rather than a plainly readable number in the captured page text, which consists mostly of the methodology surrounding the model's listing.

What the snapshot leaves open

Several details are not legible in the captured text: exact per-million-token pricing, output speed, context window size and the individual capability sub-index scores. The page text repeatedly carries a "no publicly available API" label next to the model's metrics, which may indicate the model was being tracked before a public endpoint went live, or may simply be a page control captured out of context.

The open-weights question is similarly unresolved in the text. Artificial Analysis splits its intelligence chart between open-weights and proprietary models, and the snapshot does not state which category K3.1 belongs to. That distinction matters for how significant the result is: an openly available model sitting on the frontier is a different competitive event from a hosted-only one, because it can be run anywhere at infrastructure cost rather than at provider prices.

Why it matters

If the placement holds, it indicates that the top of the capability-per-dollar curve is under pressure from a newcomer, which historically forces incumbents to respond on price, capability, or both. It also underlines how much the field now depends on independent evaluators: the ranking that made the news came from Artificial Analysis' own benchmark runs rather than vendor-reported numbers.

The caveat for anyone acting on this is that the index is versioned — v4.3.2 is the current methodology — so scores are only strictly comparable within the same version, and a frontier position is a snapshot that the next release cycle can erase.