In its video review of GPT-6 Astra, Nate Herk | AI Automation found Codex’s Ultrafast mode genuinely fast and preferred the showreel it made to the standard mode’s edit. Weekly usage was the deciding trade-off: the channel would probably turn to Ultrafast when it really needed the speed, not make it the default.
The channel compared the modes on a video-editing task and a resource guide. Ultrafast delivered the better edit in its view, but consumed much more of the weekly allowance. On the guide, it finished sooner without producing a meaningfully better result. The video describes Ultrafast as a feature of a $500-a-month plan that is also available through the API.
A 54-minute call summarized in 19 seconds
The first task was to find and summarize a 54-minute community Q&A call. GPT-6 Astra returned the summary in 19 seconds. Nate Herk | AI Automation called Ultrafast really fast.
The channel then ran the same two longer assignments in standard and Ultrafast modes. For those comparisons, it looked at both the finished work and the time and share of weekly usage each mode required. The assignments produced different answers to the channel’s practical question: when was the speed worth spending more of its allowance?
Ultrafast’s Google showreel told the better story
The video-editing prompt asked for a 60-second showreel in a vertical format. The material included two files with 11 minutes of the creator talking about a trip to Google in San Jose, plus a folder of raw footage described as more than 145 GB. Both modes received the same task, with extra-high reasoning effort.
The standard run took 50 minutes and used 4% of the weekly allowance. Ultrafast took about 19 minutes and used 17%. Nate Herk | AI Automation described that as about 2.5 times less time for Ultrafast on this task, short of the eightfold speed figure discussed in the video’s explanation of the mode.
The channel found the 17% usage cost too high for the time saved. Yet it preferred the finished Ultrafast edit: that version felt more Google-branded and told a better story about the trip. Speed was not the only difference it noticed between the two results.
In the standard edit, speech stopped before sentences were finished. The channel found those cuts unnatural. Despite the longer run, Nate Herk | AI Automation said the result felt like something GPT-6 Sol could have produced, or like work from a worse model.
The channel was very impressed that GPT-6 Astra could work through the video material and shape it into a story. It gave that reaction to the editing task while favoring Ultrafast’s version of the showreel. The weekly usage remained a concern, but between the two edits, Ultrafast’s told the better story in the channel’s view.
Both resource guides earned high marks
For the other head-to-head task, the channel asked each mode to create a YouTube resource guide based on a video about OpenAI Dots and Muse Spark 1.3. Both finished guides had seven pages and similar structures. Here the speed difference was substantial, while the channel’s ratings of the two results were close.
Free newsletter
Get the morning briefing
Start each day with the AI stories that matter and why — a short, free email from our newsroom.
Ultrafast completed the guide in 1 minute 43 seconds and used 3% of the weekly allowance. Standard mode took 13 minutes and used 1%. Nate Herk | AI Automation found no meaningful quality difference between the finished guides despite the gap in completion times.
The channel rated both results around 9 out of 10. It thought the standard guide might even be slightly better because it included useful links absent from the Ultrafast version.
That judgment was specific to the guide comparison. In the video task, the channel preferred what Ultrafast made; in the guide task, it gave both outputs high marks and saw little reason to pay the usage cost for the faster one.
Ultrafast is for the time crunch
Nate Herk | AI Automation said it would consider Ultrafast for small tasks when pressed for time. It also said the mode might make sense for people who regularly reach their usage limits. For the channel, though, the higher usage became harder to accept as tasks grew more complex.
The creator mentioned using fast mode. Separately, the channel said its work in Codex or Claude Code generally does not need to finish so quickly, particularly when the work is complex. It prefers to let those tasks run and then iterate on the results.
Its choice between modes comes down to urgency and the weekly allowance. Nate Herk | AI Automation would probably use GPT-6 Astra Ultrafast when it genuinely needed the time saving. It would probably use Ultrafast only when it really needed the speed.