On 2028-06-30, how many organizations will have an independently pretrained model scoring within five points of the top score on the Artificial Analysis Intelligence Index?

rationale:

The final projection for the number of organizations within five points of the Artificial Analysis Intelligence Index top score on June 30, 2028, is a median of 3.6, with an 80% interval of 1.8 to 6.2. As of October 16, 2026, four organizations sit within the five-point band: Anthropic (58), Meta (56), OpenAI (53), and Google (53) LLM Leaderboard - Comparison of AI models from OpenAI, ... Artificial Analysis Intelligence Index v4.3.2 https://artificialanalysis.ai/leaderboards/models Artificial Analysis Intelligence Index v4.3.2.

Algorithmic efficiency improves roughly 3x annually, while the capability gap between closed and open-weights models remains narrow at 3-4 months. However, growing compute concentration will likely stretch catch-up lags, thin the competitive frontier field, and heavily restrict the upper bound . Top models from xAI and leading Chinese labs currently score around 45-46, leaving a substantial 7-12 point gap behind the frontier SpaceXAI: Models Intelligence, Performance & Price Artificial Analysis Intelligence Index October 2026 .... Broader assumptions about AI commercialization and learning architectures do not displace this structural push-and-pull between capability saturation and periodic benchmark rebasings as the fundamental driver of the count.

Projected index rebasings entirely dictate the tail boundaries. The Artificial Analysis index frequently adjusts difficulty to counteract benchmark saturation, and a v5 update with harder, held-out tests is in active development Changelog. Harder tasks mechanically spread scores, temporarily shrinking the eligible count to 2. Conversely, periods of index stagnation allow capabilities to pool near the ceiling, inflating the count to 6 or 7.

A median of 3.6 captures the equilibrium of the three deeply entrenched leaders (Anthropic, OpenAI, Google) with a rotating competitor (such as Meta, xAI, or a fast-following Chinese lab) joining less consistently as compute concentration stretches catch-up lags. The central 50% range runs from 2.4 to 4.8 organizations, while the 80% interval runs from 1.8 to 6.2. The interval correctly spans the structural timing risks: an aggressive rebase shortly before the June 2028 resolution date driving the count downward, versus benchmark saturation artificially raising it.