Private usage benchmark.
A user's private benchmark, involving matching 60 sentences to 600 video files and extracting frames, reveals significant increases in usage for Astra models. Astra Extra High, High, Medium, and Light models now consume 8-10% usage, up from 1-2% at release. The benchmark, which measures how providers handle this task, indicates a substantial change in resource consumption, with processing times ranging from 1 hour 17 minutes to 1 hour 37 minutes across different Astra tiers.
Time & source
Times shown in UTC
Display time zone: UTC
Local time zone unavailable; showing UTC.
PublishedOffset at this time: UTC+0Sep 19, 2026, 13:01 UTC
IngestedOffset at this time: UTC+0Sep 19, 2026, 14:00 UTC
- Published
- Sep 19, 2026, 13:01
- Ingested
- Sep 19, 2026, 14:00
- Source type
- Dev community
- Tier
- Community
- Source status
- Sync delayed
Tier is a per-source editorial setting, not a per-item score.
I have my own 2year old benchmark. That I do on every model + I check the usage if possible.
Astra Extra High takes 10% was 2% at release. Astra high 8% or 9% was 1%. Astra Medium 8% was 1% Astra Light 10% was about 1%.
What OpenAi is doing is fucked.
About the benchmark. I have 60 short sentences in a file. And 600 video files. AIs need to match best videos to the sentences. They get all 600 names and they have to extract three 480p frames, first middle end frame if they think the name matches the sentence good. If they think video matches they put the name next to the short sentence. If not they check as much as they want.
It was good benchmark 2 years ago today this functionality is a given but, it tells me when a provider fucks my ass.
Edit:
xHigh 1h 37min
high 1h 17min
medium 1h 20min
light 1h30min