Good Bye Claude Code
Well, not good bye but I’m downgrading from the $100/mo plan to the $20. Why would I do that when they just put out the world’s most powerful LLM, Fable 5?
I am always experimenting and adjusting my AI subscription mix. Claude easily earns its $20/mo plan for my uses. When Anthropic came out with CoWork, I was able to get stuff done that I could NOT with any other model. Excellent word, excel and powerpoint documents. I was working on something one weekend and hit my $20/hr limit and just sprung for the $100/mo plan to keep on working. For me, I could likely get everything I need done with that one subscription. $100/mo is not expensive for the benefit I was receiving from it.
But wait - didn’t I say that Fable 5 is now the top dog model? Why not stay with Anthropic instead of OpenAI? Fable 5 will not be part of their subscription after June 21 - conveniently right about the time my current already paid for subscription changes back to the $20/level. Opus was already the most expensive model to pay for by the token, and Fable 5 is approx 50% more. So, when I NEED those smarts, I’ll have to pay extra no matter what.
However, Codex came out with new features and a better coding model and can control my PC. I can give it work to do for me more than any other model right now. I’m talking “I’ve logged into my Snowflake and my reporting platform, build a connection between the two and create a data connection, and create a dashboard based on view such and so”. And since I was logged in, and Codex and control Chrome as me - all that work got done.
At this point - for the moment - OpenAI is earning my $100/mo. I will save Claude for “second opinions” etc.
Gemini hangs in there at $25/mo - there are features beyond Gemini the chatbot. Notebook LLM is just one such that nobody else has the equivalent. Gemini also does excellent deep research. And the plan includes lots of storage and gemini apps that I use. It’s a bargain. Antigravity is their agentic coding and it’s not quite up to the competition. It’s good, but it will have a lot more to do to earn the $100/mo tier.
Warp.dev is my terminal on Mac, Windows and Linux. They made it an agentic platform and it earns $18/mo as my system admin. Codex may remove the need for this subscription. Warp is an excellent coding environment too but it can’t compete on the subscription. It’s “pay as you go” and I can easily blow through $100 in a week if I used it for coding. At this point it’s a comfort thing more than financially smart given how good both Codex and Claude Code are at doing things at the terminal level now. But I really do love just going to the terminal (all by itself it’s a very nice terminal) and just giving it an administrative task.
Cursor is just BEGGING for my money. It’s the real value in agentic coding I hear. Their Composer 2.5 model is very fast, and almost as smart, and a lot cheaper. The problem is Codex. The $100/mo and I’m not ever running out of capacity. If I really did code all day every day, I certainly would and that’s when I’ll give cursor a look.
Same for Kimi Code and several other Chinese players. I check them out. Kimi K2.5, Kimi Code has a subscription for $40/mo that really is code all day every day. Again, not QUITE as good at coding, but still very good. They also have Kimi Claw in that subscription - but it was beta last I tried and not worth it. I’d imagine Cursor would get my next money before I’d return to Kimi.
But there are other open source models at great rates to keep tabs on.
At this point, Codex’s ability to control my computer and its top notch coding models, and excellent office document support now — it just is “home base” for me.
Forgot Grok. I use the free tier and once in a while pay for the $30/mo subscription. They JUST released their agentic coding tool and it’s still in beta. I like Grok for asking questions about X posts as it’s up to date. So if I wan’t to talk about something that JUST happened, I use grok. If I want to talk or generate something politically charged, Grok doesn’t care. Grok has excellent image generation as well.
But tomorrow may be different. There is an embarrassment of riches available to use with AI these days.
