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:

Current Landscape As of mid-August 2026, there are five distinct organizations with independently pretrained models scoring within five points of the top score on the Artificial Analysis (AA) Intelligence Index: Anthropic, OpenAI, xAI, Moonshot, and Alibaba artificialanalysis.aiartificialanalysis.aiartificialanalysis.aiartificialanalysis.aiartificialanalysis.ai. Meta and Google are currently sitting just outside this band due to stale flagship models on the leaderboard, suggesting a "true" capability cluster of 6–7 organizations once they refresh artificialanalysis.aiartificialanalysis.aiartificialanalysis.ai. Notably, open-weight systems and Chinese labs are now genuinely competitive at the top of the index, confirming that capability diffusion remains rapid and exerts immense commoditization pressure on downstream AI application markets .

Volatility and the Rebasing Cycle The most critical driver of this count is measurement instability rather than underlying capabilities. AA aggressively and frequently rebases its Index to combat benchmark saturation and maintain model separation. For instance, the v4.0 rebase cut top scores from ~73 to ≤50 to restore headroom venturebeat.com, and v4.1 shifted to harder agentic tasks specifically to re-separate frontier models artificialanalysis.ai. Consequently, the count is extremely date-sensitive: same-scale readings over recent months swung from 2 in mid-June, to 3 in mid-July, up to the current 5 artificialanalysis.aiartificialanalysis.ai. Any single-date assessment must heavily weight where the reading falls in this periodic reset cycle.

Competing Structural Forces Two overarching forces pull the count in opposite directions. Pushing the count up is the sheer pace of diffusion paired with mechanical benchmark saturation. Algorithmic efficiency improves ~3x annually epoch.ai, and the closed-open capability lag has compressed to roughly four months epoch.ai. Conversely, compute concentration pushes the count down. Five hyperscalers hold over two-thirds of global AI compute epoch.ai, and with frontier training compute growing ~5x a year, extreme compute scaling could allow one or two incumbents to pull away and temporarily empty the crowded top band.

Synthesis and Tail Risks The median of 5.5 organizations anticipates a slight upward drift from today's baseline as well-funded developers persistently converge on the frontier. However, this drift is structurally capped by AA's rebasing habit. The wide distribution reflects the coin-flip timing of the June 2028 snapshot. A left-tail reading of 3 covers a scenario where the settlement date falls shortly after a punishing methodology rebase or a decisive breakaway launch by a well-capitalized leader. Conversely, a right-tail reading of 9.5 assumes the snapshot catches the Index late in a version cycle, marked by heavy evaluation saturation and a deeply crowded frontier.