Tell HN: I Hate Codex with Passion | Hacker News Reader

Tell HN: I Hate Codex with Passion

3 pointsdingdong20265 comments

OpenAI would have you believe you can do real work with Codex.

I have a $20 OpenAI subscription and a $20 Claude subscription.

With Codex I can barely get one meaningful coding task done in the 5h limit. (Either Sol 5.1 high or Astra 6 high.) With Claude I can comfortably get 3 meaningful tasks done. (Opus 5.5 high.)

This is not anecdotal but from repeated experience of dozens of tasks. Doesn't matter what's the size or complexity of those tasks - they're comparable and the differences wash out in the number. I divide my work consistently.

And now an anecdote from a moment ago: I've run out of Opus weekly usage. I've found a bug in my app. Not trivial but not really hard. I had well over 50% of the Codex 5h limit left. I thought: great, Astra will crush this bug in no time. And then I watched as the tokens evaporated and I'm left with half-assed changes. I feel like tearing my hair out. I have to wait till morning to have this solved.

Dear Sam and team: please make your product at least half as competitive as Anthropic's.

Opus is an Anthropic/Claude model, not an OAI/Codex model. I’m not sure if you can actually use an Anth model on Codex these days, but my experience to date indicates that performance tends to be best when you match the model to the harness it’s intended for.

You might also consider using the mid tier model (opus, Terra) to generate a detailed plan and then use a lower tier model (sonnet, Luna) to actually implement each task in the plan.

I’ve gotta say my experience has been the opposite. I’ve been consistently surprised at how much mileage I can get out of the $20 plan.

My usage is pretty simple. I just use codex with an AGENTS.md to point it in the right direction. I have no MCPs and use Sol on medium thinking as the main model.

For longer sessions I tell it to launch a sub agent to make changes. Otherwise I just reset my context for each new task.

Start auto-compacting your context at 100k. Your sessions will incur exponential token usage if you don't draw the line before the 1m context limit. It's the best way to efficiently use a small token quota for long-horizon tasks.

OpenCode and cheaper plans await if you can move beyond the Big Ai labs and their bluster

Um, you can put in $20 for more tokens any time.

Reply on news.ycombinator.com