GPT-6 Astra review: powerful, concise, expensive

I used GPT-6 Astra in Codex to build my personal portfolio website. After a few days with it, my impression is very positive: it is a very powerful model, and it has been very good for coding.

It is also expensive. It burns through credits quickly. I expected the cost, but that is still the main drawback of using it so far.

This is an early personal review. I have only used Astra for a few days, and my experience here is with coding in Codex.

Building my portfolio

My concrete use case was this portfolio website. I built it with Astra, so my first impression comes from using it on something I wanted to make and keep.

OpenAI's model documentation describes Astra as a model for complex reasoning, coding, computer use, research, and document creation. That is the broader product description. Coding is the part I have used enough to comment on, and even there, these are still first impressions.

I have found it very, very good. I do not yet have a long list of projects to judge it against, but building my own portfolio has given me a strong positive start.

My favourite part is the conciseness

The best part for me is how concise it is. Astra does not give me long blocks of text to read.

That might sound like a small thing to highlight in a review of a powerful model, but it matters to me. I do not like reading long responses when I am trying to get work done. An assistant that gets to the point is much more pleasant to work with.

The coding ability impressed me, but the way it communicates is what I particularly appreciate. Short, useful replies suit how I want to work.

That is my experience in Codex so far, rather than a promise about how Astra will respond in every setup or to every prompt.

The credits go quickly

Cost is the one thing that has bothered me. Astra burns through credits fast enough for it to be a clear part of my verdict, alongside how capable it is.

I usually burn through my five-hour usage allowance in 30 to 45 minutes, and certainly within 50 minutes in my experience so far. That is how quickly I reach the limit, not five hours of continuous work.

Part of the reason, I think, is that Astra is very intelligent and keeps working towards the end of a task without needing much input from me. It can get on with the work, but the usage keeps adding up while it does.

This was expected. I am not surprised that a model like this is expensive, but expecting the cost does not make it disappear.

For context, OpenAI lists standard API pricing at $10 per million input tokens and $50 per million output tokens. Those are API rates, not a calculation of what my Codex sessions cost. My observation is simply that the credits went quickly during my own use.

I would not confuse concise replies with low cost. I like having less text to read, and I still find the model expensive to use.

A strong first impression

So far, nothing else has bothered me much about Astra. It has been very good for the coding work I have done, and its conciseness is a real strength for me.

A few days is not enough for a final verdict on reliability across different projects. For now, mine is straightforward: powerful, very good to work with, and expensive.

I will update this post when I have more experience with it.