AI Battle > Rankings > GPT-6 Astra (xhigh)

GPT-6 Astra (xhigh) — Placements in 15 Rankings (2026)

GPT-6 Astra (xhigh), published by OpenAI, appears in 15 of the 129 datasets AI Battle tracks. Its strongest published placement is #1 of 20 in Terminal-Bench v4.0: Score.

Last updated: . Every placement below comes from the named public dataset; AI Battle joins them by slug and changes nothing else.

All placements for GPT-6 Astra (xhigh)

RankingRankValue
Terminal-Bench v4.0: Score#1 of 200.60
Humanity's Last Exam: Output Tokens per Task#2 of 20280
AA-Omniscience Index: Score#3 of 2043.4
MMMU-Pro: Token Usage#3 of 191000367
MMMU-Pro: Score#3 of 190.86
Humanity's Last Exam: Time per Task#4 of 201.52
Terminal-Bench v4.0: Output Tokens per Task#4 of 2017246
AA-Omniscience Index: Token Usage#4 of 20740990
AA-Omniscience Accuracy#6 of 200.62
Humanity's Last Exam: Score#7 of 200.55
Terminal-Bench v4.0: Time per Task#9 of 2013.1
AA-Omniscience Hallucination Rate#13 of 200.48
MMMU-Pro: Total Cost to Run#14 of 1910.0
Terminal-Bench v4.0: Cost per Task#17 of 200.00
Humanity's Last Exam: Cost per Task#20 of 200.00

Price vs intelligence

intelligence index →input price / 1M → $4.00
Each dot is a model, joined by AI Battle from two public datasets — the intelligence index ranking and the published input pricing ranking. Every published value is shown as-is, no averaging. The dashed line is the value frontier: no model above or to the right of it offers a lower input price at its intelligence level.

How to read this profile

Each row is a separate measurement in a separate dataset — the position reflects the metric of that ranking, and a higher number is not "better" for every dataset (times and prices are costs). AI Battle deliberately does not average placements into a single score: datasets differ in coverage, so a total would hide more than it shows.

Raw data

Machine-readable profile: /data/models/gpt-6-astra-xhigh.json.

Method

Placements, joins and this page are computed by AI Battle from public datasets — see the methodology for sources, limitations and corrections.