Pinned Tweet
๐จ Now live in Google AI Studio!
Nano Banana 2 (gemini 3.1 flash image preview)
LFG!
English
Tim Tiefenbach
5.6K posts
Tim Tiefenbach
@TimTeaFan
CX Data Scientist ๐ by day โ๏ธ Indie hacker by night ๐ Web & App Dev. All things AI ๐ค Also ๐ฏ๐ต ๐จโ๐ฉโ๐งโ๐ฆโ๐ต๐
Wiesbaden, Germany Joined Eylรผl 2010
483 Following2.1K Followers
@immasiddx Well the mouse has zero haters, the software is really bad though.
English
@LexnLin Curious, why use Antigravity in the first place when you actually want to use Claude?
English
Itโs always good to stick to one or two tools for some time even if ppl on twitter swear that this or that app is better. For me, I used Windsurf for the most part part of 2025, then decided to try Claude Code in Sep and it was so much better that I switched. Since Jan Iโm als using Codex and I currently enjoy using both, if one gets stuck often the other can help out.
English
@wholyv Why are you doing this, I mean using Anti Gravity. It might be a good product in the future but I think it is not yet.
English
@LLMJunky I have never heard of CursorBench tbh. Hard to believe GPT is so much better than Opus. I mean, Iโm a big Codex fan and often prefer it to Opus, but still, do other benchmarks shows this difference as well?
English
Wow. Absolutely amazing.
If we were to go by this one bench, this is on par or better than Opus 4.6, but lightning fast.
And even more amazing, its 75% cheaper than Composer-1 was at $2.50/mtoks on "normal"
Or $7.50/mtoks on fast, still a 25% reduction.
Even crazier, they reduced it by more than half of the cost of Composer 1.5, which was $17.50/mtok.
Bravo @cursor_ai!
Cursor@cursor_ai
Composer 2 is now available in Cursor.
English
@scottstts Iโd say bad is relativ. Try the same prompt and input with Claude and you will probably get something much better looking. Codex can also produce nice websites, but you have to be very specific, take some screenshots etc.
English
Important advice! I do the same. Above 256k results aren't great and you'll suffer Opus-like token burn!
0xSero@0xSero
Some things I learned this week: 1. GPT-5.4/Codex at more than 256k max tokens doesn't help and is too expensive (that's why I ran out of usage btw). The models still don't do great past 200k context and I can get basically infinite context using subagents anyway. 2. To be fair to OpenAI I did spam the hell out of it over 3 devices working in 4-10 sessions 24/7 especially with Autoresearch. This is very generous, and I love the app a lot.
English
๐จI think @steipete is referring to these โ kind of scam mails. Don't let them fool you just because it comes from an official Github repo. It's scam.
Peter Steinberger ๐ฆ@steipete
Folks, if you get crypto emails from websites claiming to be associated with openclaw, it's ALWAYS a scam. We would never do that. The project is open source and non-commercial. Use the official website. Be sceptical of folks trying to build commercial wrappers on top of it.
English
@MrBallaz 20$ will only get you chatGPT plus, but thatโs enough, Iโd go with that.
English
@Yuchenj_UW Never heard anybody saying that Codex produced better UI than Claude. Donโt get me wrong, I love Codex, but the UI it generates is meh.
Tim Tiefenbach@TimTeaFan
Me: Create a "new email message window" Codex: yes ๐ซ
English
People say Codex generates better UI than Claude Code.
I keep seeing the opposite. Claude just has better taste.
Also the Claude Code UI/UX is way better. Codex should probably just copy it. GPT-5.4-xhigh is very good at complex systems codebases and code review though.
English
@LLMJunky @dani_avila7 Iโm a Pro user and I think they dropped the distinction between Opus and other models in usage. At least they donโt list it separately as inb4.
English
@dani_avila7 its worth noting, if you're on a Pro account, this is an EXCELLENT solution because you have separate limits for Sonnet.
English
Did you know about the opusplan model in Claude Code?
/model opusplan
It's a hybrid alias that automatically uses Opus in plan mode for complex reasoning, then switches to Sonnet for execution.
Best of both worlds: Opus thinks, Sonnet builds
English
Peek individualism amplified by AI ๐ฑ
โpaulaโ@paularambles
LLM that keeps telling people to break up because itโs been trained on relationship advice subreddits
English
@kr0der It all depends on how many subagents it spawns and how big your repo is. But generally I agree. Time for Anthropic to introduce ads, so we can keep on chatting ๐
Tim Tiefenbach@TimTeaFan
Iโm not an @OpenAI fanboy, but I really didnโt like the Anthropic ads mocking them because they're so deceptive. So I made an unofficial answer to balance the score. Hard to dunk on ads when your free-tier is basically two prompts and goodbye.
English
@raywongy Had the same issue with an old iPad mini 2. should have never updated the OS.
English
So I decided to factory reset/erase my beloved 12โ MacBook (early 2015) just to get a clean install of the last version of macOS 11.7.11 Big Sur that it supports
Big mistake.
Apps that used to run (like Final Cut Pro) canโt even be redownloaded from the App Store anymore because Apple requires macOS 15 or later.
Thatโs right: Software that I paid $300 for back in 2020 (see second image with date of purchase) that ran on this little machine (yes, I edited 4K iPhone footage on it during major events like CES, MWC, and IFA back in the day) just straight up doesnโt install anymore.
Apple doesnโt even let me install the last compatible version anymore.
Thereโs nothing wrong with my 12โ MacBook. It should run all the software that used to work on it.
Kinda fucked up and very anti-consumer
English
@bradmillscan Iโm using KimiClaw and have made exactly the same experience, itโs unusable at the moment ๐ฅฒ
English
350-400 hours into OpenClaw over the last 33 days non-stop, no days off...I'm ready to quit.
My openclaw is fucking lost in the weeds every day today and it's driving me nuts.
Basic shit. I asked it to use GitHub. it has a GitHub skill. We have a GitHub SOP.
I can see it's thinking process about using skills, then narrating how the skill doesn't exist, then going and inventing ways to retrieve the capability to use GitHub from the internet.
I tell it to look in the openclaw docs for the proper skill path, it says "oops my bad, yeah it was there after all."
This is ChatGPT 5.4 with extra high thinking turned on.
I ask it to diagnose the problem only, so it goes and sees the system prompt is telling it to look at the wrong place, and it goes to GitHub and opens a GitHub issue about this 'bug' without even asking me.
What the actual fuck. 3 hours on a Sunday of trying to rewire the brain of my openclaw to do default-behaviour.
This thing such a productivity suck & mental poison.
I can't do anything useful or positive with OpenClaw because I'm nonstop fighting fires in the engine room.
I'm thinking about giving up.
English
drop your GPU below. i'll tell you exactly what model and config to run on it.
here's what i've tested and verified on real hardware:
RTX 3060 12GB - Qwen 3.5 9B Q4 - 50 tok/s - 128K context
RTX 3090 24GB - Qwen 3.5 27B Q4 - 35 tok/s - 300K context
RTX 3090 24GB - Qwen 3.5 35B MoE Q4 - 112 tok/s - 262K context
2x RTX 3090 - Qwen3-Coder 80B Q4 - 46 tok/s - full VRAM
all running llama.cpp with flash attention. every number is real. every config is tested. if your card isn't on this list drop it below and i'll tell you what fits.
English
@thsottiaux Really exciting to see the progress as a customer. When Codex first came out I thought I didnโt need it and the CLI was fine. How wrong I was.
English
Working at OpenAI is fun because questioning everything and taking risks is part of the culture. Within Codex, the team asks itself how we could make it an order of magnitude better every few months and then sets most things aside to go and do it across the entire stack. Some examples were the Codex App and our first deployment of Cerebras inference with WebSockets. We are now well under way on the next bet and itโs making even our best engineers nervous as itโs at the edge of whatโs possible today.
English
@wholyv What? Windows laptops cannot do this in 2026? Youโre kidding, right? Right?
English
Just had a second issue today with my macOS app where Claude one-shotted a solution implementing minimum sizes for columns in a NSSplitView, where I was going in circles with GPT. Interestingly, I was working on this problem about 6 weeks ago also with Claude and GPT 5.3 codex and back then neither of the two was able to solve it. Seems like Opus 4.6 is also stronger now in macOS development.
Tim Tiefenbach@TimTeaFan
Since I updated to macOS 26.3.1 I also needed to adjust the design of my email client app I'm working on to Liquid Glass and it was the first major problem for GPT 5.4 (high) that I bumped into. Switched to Claude that I rarely use these days and it came up with a really good analysis of why all the approaches GPT was taking did not work. It's really good to have both subscriptions. When you hit a wall with one, just use the other.
English
@kimmonismus idk Iโm enjoying it a lot too, but I also think it is very exhausting. The fact that you can now โwork" or coordinate multiple projects / issues at the same time is exhausting tbh.
English
Skill issue.
I have never enjoyed working more in my life than I do now with the help of AI.
English