Unconfirmed: posts on X say Artificial Analysis updated its Intelligence Index to a v5 ranking that makes open-weight models look materially weaker relative to closed models, but Automatica has not confirmed the change or the underlying results.
On 28 October, @NeoAIForecast wrote2 that:
What is this.. Artificial Analysis just updated Intelligence Index v5, and surprise, they make open weights look worse The gap to the leader went from 3 points in v4.3 to 12 in v5 Their top “open” model is Kimi K3.1 Pro at 46, which doesn’t even have released weights
Later the same day, @emollick wrote1:
Artificial Analysis finally moves to closed benchmarks (goodbye GDPval-AA), and finds that open weights models are falling further behind than they had previously measured. Take with a grain of salt, but matches the general vibes that we haven't seen a Fable-class open model.
Those posts agree on the direction of the claimed change — that a new Artificial Analysis version paints a weaker picture for open models — but they are still social posts, not independent verification. The available evidence provided for this report includes no primary Artificial Analysis document, changelog, methodology note, benchmark card, or ranking page confirming an "Intelligence Index v5," the cited scores, or the benchmark changes the posts describe.
That matters because benchmark revisions can move rankings for reasons that are not interchangeable: model versions may change, test sets may change, scoring weights may change, and eligibility rules around what counts as "open" may change. From the posts alone, readers can verify only that two X accounts said those things on 28 October, not that the benchmark update occurred exactly as described or that the resulting comparison is methodologically sound.
What would confirm the rumour is straightforward: a primary Artificial Analysis release showing the v5 index, its methodology, the benchmark set used, the model versions evaluated, and the scores or gaps cited in the posts. What would kill it would be a direct correction from Artificial Analysis, or a published methodology showing that the quoted claims about benchmark changes or open-model standings are inaccurate. As of the material in the available evidence, no company statement is included from Artificial Analysis or from any model provider named in the posts.
There is also no earlier Automatica Press reporting on this specific claim in the available evidence. The only Automatica links supplied here concern unrelated topics, so they do not provide background for this rumour and cannot support it independently.
For now, the narrow, supportable takeaway is that qualifying X accounts are circulating a claim that Artificial Analysis v5 worsens the apparent standing of open models versus closed ones. The posts themselves also contain their own caution: @emollick explicitly added1, "Take with a grain of salt."