[Charlie Sheen]

I have had the $100 plan for Codex for a couple months, and recently I was going to upgrade to the $200 plan for more usage, however OpenAI temporarily removed the $200 plan for the month of September then brought it back on half the amount of usage allowed.

Because of that, I decided to finally check out Claude Code. I purchased the $100 plan and started giving it some of the same projects to compare the two.

Codex is very simple, give it a prompt and it gets to work with minimum guardrails. However Codex doesn't do the very best job regardless of how thorough my prompt is or which model I use it on.

Claude takes an idea and runs with it. It is incredibly thorough. It plans out every function you would need, even if it's not in the prompt. The quality of work is outstanding.

Another thing to take into consideration is time. Codex can knock out a project in 30 minutes. Claude sometimes will take hours. However once again, Claude is very thorough and does excellent work. Codex just knocks it out, and if you forgot to add something in the prompt it won't add it on it's own.

The downside to Claude is the amount of rules and guardrails it has. Some of them I find so stupid, it doesn't want to enter any passwords anywhere, which is frustrating as I sometimes need it to login to one of my servers and if it gets logged out while working, it will stop everything until I come back and log it back in. Just an example.

Wondering what everyone else's views are on these two.

[Terry A. Davis]

I like Claude, I use their normal budget $20 plan.

I get a fair bit done, I don't mind context hitting during the day, since I use it as a way to stop working - it's kind of like peer review with other engineers, except you don't have to be nearly as people-ey.

I enjoy Claude. Sure some guardrails bite you, but what I work on it's not a huge issue.

[Charlie Sheen]

I like Claude, I use their normal budget $20 plan.

I get a fair bit done, I don't mind context hitting during the day, since I use it as a way to stop working - it's kind of like peer review with other engineers, except you don't have to be nearly as people-ey.

I enjoy Claude. Sure some guardrails bite you, but what I work on it's not a huge issue.

From what I have read Codex has an internal goal of using as little tokens as possible, where as Claude has no goal of conserving any tokens.

[Terry A. Davis]

From what I have read Codex has an internal goal of using as little tokens as possible, where as Claude has no goal of conserving any tokens.

I've not heard that, but that would align with what I see about Cluade being able hold higher context windows.

[Charlie Sheen]

I've not heard that, but that would align with what I see about Cluade being able hold higher context windows.

Have you tried Codex at all? Very little hand holding, and it gets the job done albeit not as quality as Claude.

[Terry A. Davis]

Have you tried Codex at all? Very little hand holding, and it gets the job done albeit not as quality as Claude.

I got put off GPT with how their LLM speaks, can't explain it, just really did my head in but don't feel the same with Claude

[Charlie Sheen]

I got put off GPT with how their LLM speaks, can't explain it, just really did my head in but don't feel the same with Claude

Was it the constant "you're right thats not the only way we to do this"?

[Terry A. Davis]

Was it the constant "you're right thats not the only way we to do this"?

You're right to push back on this

Just I dunno it read like it was being condescending
I haven't used it in so long I no longer remember, I only use it to generate shitty images for slideshows at work to demonstrate upcoming features etc

[Bastille]

I'm a ChatGPT user since the beginning, with the 20$ plan. I've never used Codex but I'm thinking about, besides I've unlocked the Daybreak Blue feature as my usage of ChatGPT was fully cybersecurity / white hat oriented and I don't think I could have the same thing on Claude.

I was thinking about using Codex for some projects such as mods on Minecraft because I host a server I play on with my friends, running cybersecurity audit on my homelab, and maybe for things like checking and negotiating second-hand products, especially computers that could be interesting to buy and repair, but I don't know if I could trust AI for that (I would ask it to consider the price of the same working machine, how difficult it could be to repair, the price of the parts, or if I should buy it only to retrieve the components such as SSD disks and RAM).

[Charlie Sheen]

You're right to push back on this

Just I dunno it read like it was being condescending
I haven't used it in so long I no longer remember, I only use it to generate shitty images for slideshows at work to demonstrate upcoming features etc

That quote gave me PTSD lmao.

I'm a ChatGPT user since the beginning, with the 20$ plan. I've never used Codex but I'm thinking about, besides I've unlocked the Daybreak Blue feature as my usage of ChatGPT was fully cybersecurity / white hat oriented and I don't think I could have the same thing on Claude.

I was thinking about using Codex for some projects such as mods on Minecraft because I host a server I play on with my friends, running cybersecurity audit on my homelab, and maybe for things like checking and negotiating second-hand products, especially computers that could be interesting to buy and repair, but I don't know if I could trust AI for that (I would ask it to consider the price of the same working machine, how difficult it could be to repair, the price of the parts, or if I should buy it only to retrieve the components such as SSD disks and RAM).

Muse which is Meta's AI. That one can patrol Facebook Marketplace and find items you are looking for. I don't know the entire power of it, if it will negotiate for you or just notify you. But I saw that ad today. I have a referral code for that one for 1b tokens if you want. I'm not sure if you guys have Facebook Marketplace over there. My wife just started using Muse and she likes it.

Codex is amazing primarily for code review and building. I know that you are pretty heavy into the whitehat/cybersecurity world and I'm wondering how much push back you get from AI when you're using it for your work. I know there's times where I'm just asking about something trying to get a better understanding and it thinks I'm trying to jailbreak prompt it or something.

[Fish]

I love Claude. Use to use chat, recently switched to Claud and wow, the difference is noticeable. It actually thinks about stuff before responding

[Charlie Sheen]

I love Claude. Use to use chat, recently switched to Claud and wow, the difference is noticeable. It actually thinks about stuff before responding

Yeah there were a few times where I missed a detail in a prompt, but it would've been a necessary item needed for the project and afterwards while bug testing I notice something it added and I'm just blown away by it'

[Infinity.]

I use Kimi :) kimi .ai

have used various other models like Opus 4.xx, glm 5.3 etc but Kimi K3 won the heart <3

PS: Its a little slow comparatively but does not hallucinate at times like gpt 5.xx and opus 4.xx

[Charlie Sheen]

I use Kimi :) kimi .ai

have used various other models like Opus 4.xx, glm 5.3 etc but Kimi K3 won the heart <3

PS: Its a little slow comparatively but does not hallucinate at times like gpt 5.xx and opus 4.xx

I remember when Kimi came out and everyone on here was saying it was going to be the new flagship AI and to get in on it early.

[Hern]

If cost was not an obstacle, which would you prefer for the month of October then?
I know so many are limiting usage and such like you mentioned in September so it is always needing to figure out which new location to move to and gets annoying.