← All Articles
An empty auditorium facing a lit podium and a blank screen.

NexusTrade · AI Models

No, nobody cares about GPT 6.2 Astra.

I tried OpenAI's new flagship, noticed nothing, and downgraded Codex from $200 to $100 a month.

Austin Starks Austin Starks ✦ Founder, NexusTrade ✦ November 18, 2026 ✦ 5 min read

OpenAI has a new flagship model. When I heard, I shrugged and moved along.

I'm not alone.

All over X and LinkedIn, people are calling this release "lackluster." The very first post I saw about it on X was this one.

Theo's X article about GPT-6.2 Astra, showing 1.6M views and his statement that he would not choose the model over Opus 5.5
Source: Theo on X, November 17, 2026.

So I decided to try it out. So far, I'm not impressed.

I don't notice a difference

I gave it my normal work. I had it pull the App Store analytics for the new NexusTrade iPhone app, push an app update live, and pitch me a marketing plan. I also asked it to redesign NexusTrain, the lifting and calorie tracker I've been building on the side.

The current NexusTrain homepage with a centered headline reading Transform Your Strength Training with Intelligent Programming
Before: the NexusTrain homepage that's live today.
The redesigned NexusTrain homepage with the headline Strength training that adapts beside an example bench press workout card
After: the redesign it produced, running locally.

I have two gripes.

1. The quality feels the same. I just don't notice a difference. The tasks got done, and none of it was better than what I got from the last model. The NexusTrain work was half decent at best.

2. The Codex app feels buggy. It started asking me for permission before tool calls it used to just run. I don't know what changed, and I'm tired of approving things it never asked about before.

So I downgraded Codex from the $200 plan to the $100 plan. Usually when I do that, I upgrade back almost instantly. This time I'm genuinely cutting back on how much I use ChatGPT.

Which brings me to the tier list.

Readers of this article are increasingly testing its strategy themselves.

Copy it to a free paper account, change its rules, and see if your version does better.

Test this strategy free

Free · paper money · nothing trades for real

How I rank the AI companies today

With the AI space moving this fast, I feel like most people don't understand how far ahead Claude Code is compared to Codex. Some people still think Cursor is a genuine competitor! Here's how I rank all of the AI companies right now.

1. Anthropic, and there is no second place

Claude Code gives me more usage and higher quality than anything else I have tried. The Anthropic models more or less read my mind and do what needs to be done.

I get more value from stacking multiple Anthropic subscriptions than from experimenting with other tools. Going cheap costs me more time, more money, and more bugs in production.

2. OpenAI, falling

Three months ago Codex sat in S tier right beside Claude Code. OpenAI's models were on par with Anthropic's, Codex was the best agentic IDE, and its web search worked in almost every situation.

Then OpenAI cut the usage, and now the app has started acting buggy. Codex is B tier now, and it's sliding.

3. Meta, rising

Meta is the one company in this ranking moving up. Muse Code is the choice when you are on a strict budget, and Muse Wonder made the value a little better.

It still falls short of the state of the art, and production software deserves the state of the art. For now it is D tier with the arrow pointing up.

4. Cursor, ZCode, and the Chinese shitware

The final board from my November 17 coding-agent tier list.

Six to twelve months ago, Cursor was the best tool I had. I spent hundreds of dollars a month on it. It is now the only agentic coding tool I have completely cancelled. It is expensive, the usage is atrocious, and it has fallen behind Claude Code and Codex.

I used ZCode for a while when I ran out of usage everywhere else. I learned that I would rather buy another subscription to a tool that works.

The harness is everything

Look at why Codex fell on that board. The model inside it is fine. Everything around the model got worse.

First, OpenAI cut the usage. Once I knew that I could barely use 3 agents, I stopped reaching for Codex first. Every time I open the app, I'm doing math on whether a task is worth burning my allowance. With Claude Code, I just start working.

Now the app is asking me to approve tool calls it used to run on its own. Codex is supposed to be the agent I hand work to and walk away from. If I have to sit there clicking approve, I'm the one doing the babysitting. That bug gives me even less reason to open it.

A smarter model fixes neither problem. Codex feels weaker today than it did three months ago, with a brand new flagship inside it. I already dropped from $200 to $100 a month. If these bugs stick around, the next move is a further downgrade or cancelling outright.

It doesn't matter how good GPT 6.2 Astra is. Anthropic is winning the AI race. Think about that before the IPO.

Use whichever agent you want. Then make it trade.

It doesn't matter if you use Codex or Claude Code. You can deploy automated trading strategies with the NexusTrade MCP. Your agent researches the idea, backtests it, and deploys it, all from the terminal you already use.

I'm using it right now to manage my NVIDIA trade in my Robinhood account. It isn't going well.

My real Robinhood account, live. Open the full portfolio on NexusTrade.

My Robinhood account went from $17,443 at the end of October to $5,656 today. That's down 68% this month, almost all of it from five NVIDIA calls that cost me $12,045 and are worth $3,873 now.

Want to know what I do next? Follow my profile on NexusTrade and watch it happen in public.

Follow my trades on NexusTrade

See my portfolios, the strategies behind them, and every move I make on the NVIDIA bet.

Follow Austin

Readers of this article are increasingly testing its strategy themselves.

Copy it to a free paper account, change its rules, and see if your version does better.

Test this strategy free

Free · paper money · nothing trades for real

Put your research to work

Create a free account to research a company, build a strategy with Aurora, or inspect a historical backtest. Choose the task you want to try.

or

Free to browse. No credit card required.

Discussion

Sign in or create a free account to join the discussion.

No comments yet.