I really would like to see OpenAI’s focus on efficiency but everytime I use Codex, it wastes tokens like there’s no tomorrow, hitting week limit in a day, where I’m able to use Claude just fine. Maybe it’s based on the codebase, I don’t know, but I have better results with Claude than Codex.
My boss uses opus and gets good results when I use it always burns tokens. The other way is also true too I get great results with sol/terra but my boss does not.
One of my challenges is that I often end up burning tokens trying to build the right solution rather than building a solution that is good enough and deferring the right choices until later. I have been actively working to change my expectations for building with AI to accept worse solutions to get things going rather than trying to solve everything at launch.
This has been a recurring problem for me - as a security engineer catasrophization is a fundamental skill to finding vulnerabilities in complex systems, but makes me to conservative when building. For some of my leaders they are much better at saying 'good enough, ship it, and fix it later'.
Maybe. But if Claude is fine with my prompts while Codex uses them as an excuse to burn tokens. I have nothing against the model, they are more or less on pair, it’s just that Claude is giving me more value for same money.
Codex is fabulous at work, where token use is near limitless and ultra-thorough tool use is welcome. Go ahead and fire off searches for look-alike terms on my 96-core cloud instance.
By contrast, Claude Code's bias to make assumptions of reasonableness about underlying systems has proven to be immensely frustrating over the last month or two, both personally and at work. I've wasted days on "that was my mistake. I've been reporting numbers on the old architecture because I hadn't enabled the new one in the config" both at work and home. It's immensely frustrating.
But here we are. Wrestling with energetic idiots in model form, wrangled by over-specific harnesses that struggle to stay off of deranged side-quests.
I had the opposite experience. Claude models and its harness feel like they are set to eat tokens for everything, especially if it’s ultracode effort. I have seen it spawn 6 agents and eat my 4-hour quota right in front of me. Codex is on point and follows instructions well even with max effort. I like my analogy of Claude being garrulous and Codex being laconic.
If I see more limit hits in same sitting session, I have to rearrange my workflows. my experiments with qwen and deepseek have been good, cant wait to try glm and other models.
Just before I canceled my 20x Max sub I had a day where I had ~15% of my weekly I was trying to burn. I set it on ultracode and because I had set the max agents 32 for another project and forgot, the five research agents ended up spinning up a total of 26 subagents and burned through the remainder of my weekly in the span of 20 minutes before I noticed and shut it down.
Claude did the same thing to me the other day. I ran out of my 5 hr limit, and it went into my $100 credit they had given me. Multiple parallel agents. It ate it in no time flat.
When I checked, all that credit was gone, I still wasn't into the next 5 hours, and all the agents had failed, returning nothing.
I didn't even get anything for burning all that credit. If I had paid for it, I'd be very, very pissed.
Similar experience here. I had raised my cap to $20 because they had given me promo credit. The credit expired without me using it and I didn’t notice.
One fine day I was like, oh my weekly quota resets soon, I will kick off an expensive bug hunt. It launched parallel things and burned $10 in like a moment.
Here is an excerpt from the system prompt for UltraCode (Same for Fable,Opus,Sonnet):
"Ultracode. When a system-reminder confirms ultracode is on, that opt-in is standing: author and run a workflow for every substantive task by default. The goal is the most exhaustive, correct answer you can produce — token cost is not a constraint."
