As of 15 Oct 2026, 11:38 UTC
Ten capability fields and their evidence coverage: for how many of the current frontier models (OpenAI, Anthropic, Google, SpaceXAI) is there a public benchmark result with a source in each field? The terrain shows this coverage — not how capable a model is, and not how close AGI is. There is no overall value.
Height = share of the 10 current frontier models (OpenAI, Anthropic, Google, SpaceXAI) with public benchmark evidence in this field. It shows how well a field is evidenced — not how capable or intelligent the models are, and not a distance to AGI. Frontier set checked on 4 Oct 2026. Which models?
The same data as instruments. Outer ring: one segment per frontier model, lit where the model has public benchmark evidence in the field. Middle: one dot per own run of agi.jetzt (filled = a check was decided). Inside: one diamond per work product, ochre where a capability mapped to the field is usable — that mapping is an assumption of agi.jetzt, not a measurement.
Three sets of models, each named. None of them is the registry: it also holds speech, image and embedding endpoints and retired models, which no test here was ever meant to cover.
6 of 28 runs have a decided check, 0 of 28 judge evaluations are published. Runs from 17 Jul 2026 to 1 Oct 2026. None of the 6 tested models is in the current frontier set.
Results that benchmark maintainers or providers report, each with its source. They are not averaged and not compared across benchmarks.
The models each provider itself positions as its top tier, one to three per provider. Set checked on 4 Oct 2026.
Coral dots: the 5 days on which agi.jetzt ran its own tests (28 runs). Names: releases of text models by OpenAI, Anthropic, Google, SpaceXAI; models of the current frontier set are highlighted. Speech, image, embedding, moderation and realtime endpoints are not listed. No history per field is reconstructed.