WHAT NEW IN X11.05 THU

AI collections from X · 22 collections · 67,905 posts

gpt-6IMAGE

VulcanBench: GPT-5.6 Terra nearly matches Astra at lower cost

Morgan tested five core OpenAI models on VulcanBench: Terra Max scored 89.05 against Astra Max's 89.30, both solved all 23 tasks, and Terra cost $1.88 per task versus $2.57 for Astra.

J A Z I I
@notjaziiArchived 2026.09.20Originally posted 2026.09.20
VulcanBench: GPT-5.6 Terra nearly matches Astra at lower cost
J A Z I I @notjaziiOriginally posted 2026.09.20
Open original on X

gpt 5.6 terra = gpt 6 astra morgan tested openai five core models on vulcanbench and somehow terra max almost matched astra performance > terra max: 89.05 > astra max: 89.30 > both solved all 23 tasks > terra: $1.88 per task > astra: $2.57 per task terra was just 0.25 points behind astra while costing less already switched from astra to terra, let’s see how it does but i don’t understand one thing if terra is this good, why is openai removing it from gpt 6 family?

This post quotes another post: “VulcanBench results for five OpenAI models across all effort levels”

x.com/notjazii/status/2101360203057037790Archived 2026.09.20 10:08